Topic: Pune ICAI-AI Tools – Session 17, scheduled on 27th May, 2026 at 5 pm

Añadido: 2026-05-28

[00:00:00]also so people who would want to actually revisit can visit the so uh good evening everyone uh and welcome you all after the long pause we are again starting this uh sessions on I use Yes.

[00:00:39]Okay. So uh good evening everyone again and uh thank you all for joining this session after uh uh like a gap of almost a month and uh we have amongst us uh CA prana Tulshan who is also an AI enthusiast like we have had many uh faculties in the past and uh while restarting these sessions I had in mind that uh why not We try to take a particular tool and also know a little bit about what if we go on a paid model because till now we have mostly discussed about free uh tools how to use them but uh I think we all know the importance of AI in our practice now and we can surely start using maybe a little bit of paid models. So um today we have Prana uh who will talk about MCP and connectors. So he will tell you more but yes these will be uh this will be a very important session with regards to how we can connect the various softwares or tally or other things which we maybe existing not exactly tally but generally these add as these work as a link between that existing things and AI right no >> and and uh if we if we learn this and we start using this then our life would be more easier in terms of uh taking the data from a certain place to directly to AI or working through uh through AI on these uh softwares. So um and we uh we we actually plan to uh do more of these uh for claude itself in the future and learn about different other aspects of claude like skills or maybe how claude can specifically be used in stock.

[00:02:51]>> So I thought we will start with the NCP and connectors which is a very important topic. Cool.

[00:02:58]>> Uh Pranav over to you and before we start I would like to request you to give a brief introduction and tell a few lines about yourself so the uh our audience know whom they are listening to. So uh and at the end please note your questions at the end we can take a few question and answers after he finishes his uh presentation. So over to you PR.

[00:03:25]>> Sure. Uh thank you ma'am. Thanks for the a brief background which I've given for the session and uh first of all I guess I would like to uh like thank the pun branch for taking such a important topic of MCPS like I have had many sessions I've taken multip multiple sessions on AI but usually people still cover the surfacial part on AI that how we can use it connecting tally to AI and all this uh similar stuffs but uh this is the first time that someone has actually um told that they want to learn about MCPS and all so it was really a a nice topic.

[00:03:58]Definitely it is a great topic to have and it is uh definitely something which is worth knowing.

[00:04:04]So okay I'll just uh before I start I'll just start a bit with my uh introduction. So okay so yeah my name is Pranav Tulsan and uh uh talking about my academics I've had all ranks during my academics and all. So uh before that I I had worked with PSU uh with an with a PSU I was working for 4 years. Then I also worked at MOPG for one year looking after the governance of the ministry of petroleum and natural gas.

[00:04:33]>> So as an additional director I was working there. Then since then I'm working as an independent consultant for D2C startups. Uh and usually in the finance ecosystem I myself I have a lot of interest into the AI thing. the uh the AI thing how the AI thing is impacting the finance aspect. I particularly so interested about IT security and IT compliance SAS systems and how IT infrastructure work how the CA practice and tech transformation is working and particularly around DPDP act is something of my core area of concern for the tech and finance industry. So yes, I've been teaching for last four years Python through my YouTube channel and uh and I've mentored multiple finance students over the Python and Python thing. So they definitely AI is something which is recently has come for last 2 three years. But Python was there for last 20 years before 20 years. So I started using Python in 2020 and since then I've been doing a lot of automations around it. I built document management system, client portals for CAF forms, folder utilities, GST consistencies and all. So that's my experience around the tech domain particularly. So today's topic MCPN connectors again a brilliant topic to have and since ma'am said we are looking forward to have more paid versions of cloud. So I guess we will be looking at many such capabilities of claude which definitely deserves to us for having a paid version but also at the same time since I said my more expertise towards the TPP act compliance as well. So we'll look at the risk as well what the risk clot poses in terms of on the angle of DBDP act. So starting with MCP and connectors your AI works force how they are connected to your practice.

[00:06:14]So uh I've talked about my profile and uh okay so I'll start with what are we going to cover uh today. So there's a live demo that I give that what exactly MCPR are and what are we going to learn at the end of the 60 minutes today which we have and multiple ways which we can use the data there are multiple problems in regarding uploading of data to claude and we'll see around work around around that as well how DPP act uh will affect our workings how the and then we have some MCP demos to give that how today we'll not only see the MCP working but we'll see how to create those MCPS okay because that is where the power lies It's not about I giving you a MCP, you implementing it and the work finish is there. It's about moreover you learning how MCPs work and how we can actually self build such MCPS and uh only basic knowledge of Python will be required if you are trying to build an MCP or some sort of system and then I have lastly have some resources which I'll save with the Pune branch and uh from that I guess you can benefit out of it. So, so the uh concept of MCB is like if you have your data sets with you and if you can directly ask your claude regarding that data sets and claude will answer you based on your questions and there no hardcoded questions first you can ask as per your own convenient and based on your questions claude will give you the answers. So suppose we ask that which 40 clients have GSTR mismatches more than 50,000. So we have taken a case of GSTR 2B reconciliation of purchase register and 2B because it's a very known case across everyone and everyone can relate to it. That is why I have taken this use case and suppose you ask this and we get an answer in 20 seconds. Okay. And along with a chart saying, okay, this is the these are the 40 clients who are having the misal.

[00:07:57]>> Okay. So let's see. Let's see this live.

[00:08:00]Right now I have this claude with me.

[00:08:02]I'll start a new chat and I have a few questions. Suppose I want to ask it a few questions. So suppose I ask this what is the GSTN of Rashri product lines. I I have my data set. So we'll definitely see how this is working. But let's see this in action first. So it is loading some tools and uh it said GSTR is this okay we found by an exact match okay perfect it's able to see the GST of this let's ask does the ITC reconation of this vendor tell me how much ITC is at risk it's like a very pure English I'm asking that if I try to reconcile the ITC of this vendor so first of all first of all see that I've not attached any file okay I've not attached any GST file or something there's definitely a connector attached to it which is this GSTN lookup and this is What we are going to see that how we build this connector okay how we build this MCP so I'm not attached any file I'm simply asking do the ITC reconation of this vendor mean you perform the ITC rec of this vendor and tell me how much ITC at risk means how much ITC I have claimed but the vendor has not filed okay but I didn't explain so much I simply said how much ITC is at risk like I can like I talk to a normal CEO or person who understands it so it says here's the ITC recon summary matched there are 15 ITC are matched and ITC at risk that is in your books missing in gesture 2A is one and they're not booked it gave me an additional information as well this ITC not booked that is present in gesture 2 is missing from your books to invoices ITC address is 2290 rupees and invoice number this dated this is present in your purchase but not filed by vendor since the vendor has not reported invoice you cannot claim ITC it gave some additional information as well again second not booked but missed IT opportunity You could have availed this it but you could not avail because it is not recorded in your purchase register.

[00:09:53]These are the missed it opportunity.

[00:09:54]These are the invoices. So this is what it is telling me. Okay. Let's ask some more questions. Are there any invoices where the tax amount in my books differs from GT 2A? So like suppose now it say the invoice was found. Here the invoice is found. It is matched and safe. But now I want to ask an amount level if the amount was matched but is the tax amount differing.

[00:10:16]So let's see what it tells. It is again working using some tool. Yes, there are significant mismatches. Here's the full breakdown. Then it says there are multiple mismatches. 15 in was a mismatch. So again the problem is it says 15 were matched but when I told it to match against the amount it says all the 15 are having a mismatch. So here even I doubt that whether it is correct or not. So what I'll do I'll once try to see my base files which is there and let's see the base files which I've given to it.

[00:10:50]So uh for the demo I had these two files. Okay, this is combined GST 2A.

[00:10:54]Let's open this file.

[00:10:57]So this is exactly the file which it's referring. This is a file. Okay, and uh let's look at the purchase register as well and let's try to cross check whether it is just hallucinating or it's actually giving me the right data.

[00:11:12]So we talked about rasi road lines.

[00:11:14]Okay. So here I've already filtered it.

[00:11:15]So Rasi road lines it is my purchase register. Let's keep side by side and uh I guess some uh partition participant might be have that access. Okay thank you. So let's uh compare. Okay. The one how we do the reconciliation. Uh Sumei G I guess your screen has some annotations going on. You can turn off that.

[00:11:45]Okay. So let's look at this. So I have Ras my GSTR 2A. This is my uh purchase register. Okay. Let's look at invoices.

[00:11:57]The these have 15 invoices in total.

[00:12:02]16 invoice in total. Okay. It has 16 invoices and it has total of 17 invoices. Now let's compare the amount which it was telling.

[00:12:11]So in purchase register I have amounts like total taxes the IGST is there I have this 1377 2390 and all but if I look at this is my 2A if I look at the 2A it has filed at zero that's a concern for me because it was though it was showing as batching but it says the client has filed as zero it may be that it is a road lines and maybe RCM was applicable but it didn't file at RCM it supply track RCM said no but still it filed as zero so that's a concern for us to raise and that's what claude correctly pointed out that okay the amount if you see there all these amount are having mismatches that 19 number invoice has zero GSTR in 2A but in books I have 1377 if if I cross check this then 19 number 1377 is the 2A I have and 19 number is the invoice and if I see 90 number over here in two way there's no at then there's no tax in any of the invoice okay I just forgot to confirm I hope the English is fine or we should mix with Hindi as well.

[00:13:14]The language is fine for any participant can confirm if they're able to understand able to follow through any confirmation in the chat.

[00:13:29]Okay. Yes, please continue. Thanks.

[00:13:31]Thanks, Sachin. So we have the Smith math and u so Claude was right basically that there's no ITC in Jester too but going in your books also if I see invoice number 20 it was saying ITC at risk that is present in my purchase register but not filed by the vendor. So let's see invoice number 20 was there and here if you see in 2A there was no invoice number 20 correct.

[00:13:54]So that that was one this invoice number 20 the ITC was at risk. Second it gave me two invoices which were there in 2B 960 and 989 which are not the purchase register that's the missed ITC opportunity. So you can see these two invoices were there 960 and 960 were not here. So that is exactly how the reconciliation was done by claude and we can now verify it was correct. Let me ask some more tougher questions to him.

[00:14:20]So if I just type give me a monthwise ITC summary. So let's see whether it's able to give me a monthly ITC summary or not.

[00:14:34]So you have and just look at the speed of the response. It's much faster. Okay.

[00:14:39]Like uh had the data been uploaded to claude it will be much slower. If anyone have tried uploading data of such high volume, it will much slower. So look look at the data also. It will be like matching exactly and there's no chance that it will be doing any kind of hallucination because I'm using MCP for it and not a typical uploading to cloud.

[00:14:59]Okay. So if I look at total tax in gesture 2A it's coming as 3 413 69745 3 413 69 745 and same for purchase it will be working. It also gave me key observation in July, September and total mismatch. It gave me out of it own intelligence.

[00:15:18]Lastly, I ask one uh bigger question.

[00:15:21]Which vendors are both non-compliant and have high ITC risk? So, it's a great question. This requires combining compliance score with ITC. So, it has to use two or three functions to for this.

[00:15:31]So, again using that tool and here's the combined analysis vendors that are both non-compliant also and having high it amount is also high. So there are 33 non-compliant vendors and it gave me the highest priority that these are the highest vendors which are having non-compliance score of zero like they are not having any invoice matched and then gave me a percentage of some 100% but uh the amount is not matching and all. So key takeaways to give these are the small demo of how exactly MCPS help us to get uh data insights with talking to um the cloud without having to actually upload it. So a typical person using AI would have uploaded both the data sets like he must have uploaded both these excel files to cloud and must have told all these questions to go through that would work perfectly that's that's not an issue that that would have worked upload your uh data to the cloud and ask questions to it and second way is the MCP which we'll discuss but what's the issue with the way one didn't I go with the direct upload so the problem is the client data goes to entropic servers and when I say client data the entire file goes to the entropic servers and I'm not sure whether you know of the entropic data retention policy or not but if you have um uh ticked mark on the help me improve button if I so you have ticked mark on the help me improve button then your data will be stored with clot for 5 years and the data means the data will be there at clots clot server along with all the things which have uploaded so that's the first DPDP risk second is context limit If you keep such large file as input context, your tokens are going to expire and your tokens are going to expire and that will again lead to your higher cost definitely. Second, third is the highest risk which I believe h hallucination risk because you don't have control about the logics. It does not know what it is risk means. So you don't have any control of the logic. It is AI is probabilistic and not deterministic. So that is why that becomes a concern many times that are not able to have the control of how the logic will be applied to answer that question and then slow definitely it will much slower than the one which we saw now why MCP wins every time data never leaves your system data is there in my Excel file it's there in my this thing it's there inside of my D drive and this ma'am if you can ma'am you can mute all option is there if you can mute to everyone. Okay. Data is already there in my uh systems and it can system near zero. There cannot be any hallucination in terms of response.

[00:18:06]The comments which it gives there might be some hence otherwise this is like pure 0% chance of hallucination. It can handle 1 million rows or 10 million rows because processing happens in your local system. Okay. It does not happen through any uh in any cloud. So if you look at this I have this pkl file with a pickle file. Pickle file is a more database indexed file which is faster in terms of read and writes. So we use pickle file.

[00:18:32]This is autocreated by Python codes and that is how the read and write are so faster. So it can handle any one/10enth of the token cost. The token cost is one10th literally one/10th because none of the data is uploaded to the cloud server. Whatever the file the files may be 1 MB, 5 MB or 10 MB. It will always stay in your system only. And yeah logic stays in your hand. You have the control of how to write the logic because you write the Python code and unlimited scalability. That's the benefit. So this is a head toe control that DPDP compliance is much compliant in case of MCP context limit hallucination token cost response speed my control and scality. In all the case MCP wins against direct uploading my data to cloud. So again some aspect of DPDP act that uh 2023 the DP act was passed by the parliament. January 25 draft rules released. 2026 rules are being finalized and many rules have already been notified and 2027 full enforcement is planned. But what does it mean for the CFM like us? So it has a word called data fiduciary and uh CFMs any data fiduciary is any person who holds financial or sensitive personal data related to any data principle which is a client. So since CFM already hold much much of financial and critical and very sensitive data for the clients they are falling well within the terms of data fiduciary and so being they being a data fiduciary they have to act as per the DPP act because they are holding the digital personal data of the clients.

[00:20:02]How claude stores your data as I already said uh Claude stores your data for 5 years and uh let me see if I can give that demo. I have a session tomorrow for the DPDP act specifically but let's see if I can give you some glimpse of it today no Okay.

[00:20:42]So this is a um suppose when you suppose we do a first chat of when you first time chat with claude or when you first first make your account you get an option like this before your first chat a few things to know. This is what claude ask you here. If you see in this particular uh space it says help claude improve allow the use of your chats and coding sessions to train and improve the claude change anytime in privacy settings. So it says whether you want to help or not. So I did I don't want to help. So now let's see what does it mean when you say that I want to help claude improve it's the data is stored up to 5 years when you obtain. Now this answer is being given by claude itself. Okay, I asked basically cloud only what does it mean and I just asked simply what about the storage of data Excel files and chat how how do you store this? So it said that if you opted in I would have stored the data for 5 years and I I would have used it for training when it says in deidentified form it will train but many times the human eye will see your data.

[00:21:41]There's some humans sitting in claude's office and tropics office and they they can see your data. Okay. If you have opted if you opted out still they will hold data for 30 days. Okay. But after 30 days they won't use for training.

[00:21:51]They will delete the data. Now remember it does not mean that your chats will get deleted. Okay. Your chats will stay with you as long as you want them to stay. But the data in their server for training the the training server which they use at least there your data will not be stored. So these are something that you should have and uh yeah so basically it has it has speaking that only that when you use cloud it is important that to know how cloud is storing the data. There's a lot of nuances to it but yeah this is the basic thing that if you should not be uploading many financial data or compliance data for the cloud the penalty if you ask for the penalty for the DPTP act it's not small penalty as big as 250 crores this is a big big sum and definitely it looks very not natural that such a big sum of penalty being applied and the act is being governed being covering CAMS as well but that is the reality that in in the first instance it looks that this penalty is there and if we don't take reasonable safeguards it may imply to the CFMS as well. So what does the DBT compliance actually say? So just a small that it says they were obtaining the client consent, verifying the country where the data is stored, confirming how long data is used by the client and signing a DPA with the SAS with the with the provider breach whether which notification SO be documented or not. Pseudo anonym uh pseudo anon anonymization is basically hiding the personal data and making it not known identifiable to a particular person. client portal mechanism backup data is there raw client data never posted into a chat so when we look at at least MCP then MCP is helping you with some of the compliance and what other compliance can be done there are ways to fix it it does not mean that we don't use cloud definitely we can use cloud and we can still comply with DPB act as well but that's a topic for separate day today let's speak to let's stick to MCP thing only how heavy is the hammer so basically 250 cr of penalty if if we have failure to implement security safeguards and that leads to a data breach then the penalty can be there and all these penalties are there for different types of non-compliances but now let's come to our thing what is the MCP the simple mental model to understand MCP is that when a claude is a brain okay claude is a brain that can that is very good with understanding text now MCP server what it does it now when I have a excel files I know that when I say that okay compare do a reconation I'm simply mean that take this GST in of this Rasi road lines or anyone any any vendor take this GSTN use this GSTN to search my purchase register and then search my GSTR2A and based on this does see what invoices are matching or what are not matching so if you see this does not require a brain it is a simple Excel or arithmetic it's a proper rule-based calculation so MCP does that thing MCP does this calculation inside of your uh inside your laptop only before giving the response to claude. So what happens? Claude will see the question. What question is being asked by the by the user? The user ask which client have to be mismatch more than 50,000 B in the plain English. This is being done in plain English. Now the claude is a brain which will write which will uh which will understand which tool to use which Python function to use to get this data and based on that Python function that will be run in the MCP server means MCP server basically your local system and then the output of that server output of that Python code will be given to claude as a response. Cloud will not know what the MCP server did.

[00:25:35]MCP server is proper internal working.

[00:25:38]It will just say okay this question is there run this function and then based on the function run it will return something that claude will take and it will give to the user as a response that is what cloud so there is a even though my data might have 10 lakh or 20 lakh rows but if the uh logic which is applied if it is returning only 10 rows only 10 rows of data context will be applicable the context of 10 rows of data like a smaller tokens will be applicable so that's what it is so I'll Let's move to demo and I've made three demos beginner and intermed. So we'll go one by one.

[00:26:15]Okay. So let's go at demo first. Uh uh anyway anyone having any questions they may hold it for some time right now and yes I will definitely take up all of those at the end.

[00:26:41]Okay, first tell me I just want to know in chat anyone have used Python before and anyone has any idea of Python? Okay, sorry to sorry I'm writing I'll zoom the PT now.

[00:26:51]Okay to Hindi mix.

[00:26:57]Okay. Okay. Okay. Okay.

[00:27:00]Anyone have?

[00:27:07]No. No. Yes. Yes. Yes. Okay. Some. Yes.

[00:27:09]Used it for few automation working tax.

[00:27:11]Great. Yes. Okay. Thanks. Who is this?

[00:27:14]DN DPNS. Thank you. We have started using that's great. That's great. Uh miss. Yeah. Right. So um basically when we talk of this uh MCP server thing so I'll I'll I just had three levels level zero level one level and then the final level. So now look Python is very what we say earlier when like I started in 2021 that time it was very harder to learn Python in terms of that we had to learn each each and every syntax of it how to relate to our use case and each like all this 100 lines of code this used to be handwritten okay there's no copy pasting as such we have to literally type here that okay from mcps server and this is how you progress writing the code but now things have changed drastically as you all know and now there's no need to write all this code. Even the like this this setup or the demo which I have to do for you that demo I did it maybe I asked this chat GPTA this claude to only write it and I use cloud inside of my this is a VS code this is a like a code writing platform and I if you see I I use I use cloud code over here I use Gemini I use open code editor I use uh this is the copilot of GitHub I use continue and I and I use blackbox so there are like five AI tools who help me write the code when the token of this get expired I use that and then the context already there so it's very easy to write code that's the baseline I want to say it's it's have not been uh difficult it was difficult definitely earlier and Python has anyways been the easiest language to learn so I guess this is a good start if anyone who is new to it can learn it so so those who are looking at Python for the first time they might be a bit like what screen is this but uh let me uh see if I can explain this okay so the All this blue thing is just a comment. Just see it's a word file where something is written. This is a possible simplest possible MCP server. No tool, no external files, no database, just pure Python logic. What is an MCP tool? MCP is a Python function that Claude can call on your behalf when you ask it the question. So Claude will call a function on my behalf. I will ask the question and Claude will determine which function to call. That's the first thing. Okay?

[00:29:26]Because we have multiple functions inside of a Python code. Like if I say this then I have all these are function def nom invoice def build build keys or like load with cast let me see if I can have better names yeah get gsti so like when I asked the first question the first tool give the gsti of this particular vendor then it it understood I have to call this particular code so because look this is the triple dot whatever is written here is a pure English only it's pure English it's not a code so look I write already I in my python code I already write it.

[00:30:01]>> Screen is not visible. Sir, >> my screen not sorry my screen is not visible.

[00:30:07]I was showing my screen. I thought you all are looking at my screen. Thanks.

[00:30:11]Thanks for telling. You would have continued like this only otherwise.

[00:30:14]Okay. Sorry.

[00:30:18]Okay. I'll I'll start a bit.

[00:30:21]So this is a VS code screen. Um so how does it work? It's simple like this is my normal file explorer. This is screen we always look at. Now when we open a VS code here like open with code if I do this then this screen gets opened. Okay.

[00:30:35]When I look at this screen this is nothing but this that the all these files I get uh to see easily that now when I have to look at from here I have to double click and open it. Right now the VS code what it does for me it uh it simply helps me to navigate through multiple files. That's the VS code I was saying about the tools which I use in VS code. It was all about Gemini. Gemini code I use here's the open codeex uh editor which I have this is a GitHub copilot which I use and I use this black box is there which helps me to write the code in a much faster manner and and there we have continue.dev the continue is again a agent orchestrator which combines which which controls all these different agents and helps me write the code. So I was trying to say this that writing a code is not a big issue and even this code which I'm showing to you it was written by claude only and look at this this is the chat which I did with claude to write all this code okay so it's very like a simple thing no one is writing code from scratch nowadays okay it was 2023 people used to write after coming clude no one is writing code so only the agents are writing the code so I was just showing you three levels of uh MCPS are there like this MCP whenever you uh see this uh GST intelligence suit MCB server. So I have these six tools made. Okay. Get GST it will look up for GST by vendor name because maybe that the customer the user don't know the GST. It will ask the vendor name that give me of ABC enterprise give me of this. The first tool which I need is get GST. Okay. So that is where I load this first tool and this load tool. What the cloud does it it knows it will read this file. It will read this Python file. This has no client data. It just has logics. It is written by me and I can write any logic which I want. So I have written over here that okay this is the get GST function. Use this when a user when to call this tool when use whenever the user wants to find a vendor's GST register number. So whenever you get such as what is the GST of this vendor find the GST of this what is the for this. So I've given examples okay whenever a user gives all these examples that time you use this tool. Okay. Okay.

[00:32:45]So now it say GSTN needs a vendor name.

[00:32:48]So whatever the user gives that will come as a vendor name. Now it will run some function. Now def this function is a Python based function. We might not understand it right now but it's a very simple pandas implementation. And using the pandas we are doing this all the search and all we are doing. Okay. It will at the end will return the vendor name. It will return vendor name and the this GST number it will return. That's the function what it will do. So in this way whenever the claude gets a prompt where I'm asking what is the GST number of this it will simply come it will read the stock string it will read and it is intelligent because we know claude is intelligent enough to understand that okay this function does this the user is asking me for a GTM I have to use this particular function so this function will get called and this function will run in Python it will simply it has been given the uh data that okay you look at my uh what I say you look at my uh purchase register my GSTR 2A. Okay, this look if source code GSTR 2A then look at this this file if purchase register and look at this file. So that logic is written over here just like the M was talking about the skills. So just like the skills the logic is written that what you need to do but written in the Python language and the cloud will simply take the output from here.

[00:34:00]Similarly another tool is ITC cancellation. So whenever the user says which invoice I'm missing in 2A which vendors have not filed their returns what is my ITC at risk. Now look why could Claude understand the word ITC at risk. Now there's a multiple meanings he could have assigned but I have already given an example that these are the words that my users use. If they use any of these words any of these questions run this particular function and when this function will run it will automatically do booked notebook all the logics I have written that okay what logic I want and very simple write logic as I already said I have we have you have like three four set of AI agents to help you to write the Python code. you don't have to write any single line of code with you. Okay. So that's the one thing and uh okay so uh so okay so that was around the logic writing. So similarly I have this one tool I have the tool of vendor compliance score I have the tool of duplicate invoice check. How to do the duplicate invoice check? How to do the monthly IT summary. All these tools will return how much it is available. This this whenever these type of questions are asked this tool will get called.

[00:35:06]These are the six tools I have. So that is why whenever I used to ask the question to Claude over here.

[00:35:14]Which one? This one.

[00:35:26]I think some people are complaining.

[00:35:29]>> Okay.

[00:35:29]>> Is it visible?

[00:35:33]>> Sir, please take H now. Now now, now it should be visible. H.

[00:35:41]So I was saying these are the one Python code is the base which you have and the cloud is your brain through which you will you can do your particular thing.

[00:35:53]Okay. So let's see which huh this was the cloud. No, this was the CL I was using. Yes.

[00:36:04]Okay.

[00:36:06]So let's come to my uh this presentation fast. Okay. So as we said there were um uh like the Python code which you write that is connected to cloud. Now I'll show you one uh step by step of how how can you get this code and how these things are stored where these things are stored.

[00:36:23]Let me first stop my I'm just uh let me see the right profile where the claude um claude chat was there.

[00:36:31]Uh so it was there in this. Okay. Yeah, this is okay. It's on desktop.

[00:36:47]Okay, got it. So okay now the screen is visible. Yes.

[00:36:58]Huh. So that I was saying this is the uh claude which uh and and this works in your cloud desktop. Okay. First uh first uh thing it'll work on your cloud desktop only not on the cloud cloud because as I said uh this works with the local file right. You're not uploading your files anywhere. So you need to have that cloud desktop. So cloud desktop you can download through uh online anywhere when you write cloud it will come as a app for you. Okay basically this is how cloud comes. So this is my uh cloud is my desktop. Okay as you can see here it's in my desktop and second thing which I was saying now second is the configuration that we have to do. So where do you get such MCPS? So like I have posted this MCP in my GitHub. Now what is GitHub? GitHub is a basic platform where just like Facebook people used to post their photos and their statuses. LinkedIn people post their successes and all. So in GitHub people post their codes whatever code they write. So this is my GitHub profile and through here you can see I have multiple repositories. One is the finance MCP. I made it a public repo. So whatever code I wrote I will even in future whenever I update any MCP I will keep it over here. And since it's a open so anyone else can also if suppose someone else has also written an MCP they can come and contribute to this repo saying okay I have this particular repo and people can come and find repos here. That's a very simple logic of having a GitHub repo. So I have this GST RECO thing and inside GSTO I have this GSTO MC file. This is a single file.

[00:38:31]This is like small file which you need.

[00:38:33]This file you can download into your system. You can keep it uh anywhere you want. You can keep it any place. And then I have told how to use this. Okay.

[00:38:42]GST intelligence MCP for CAF forms a plug-and-play MCP which connects cloud desktop directly to your GST data files.

[00:38:48]So this is the GST record. I was explaining how to do this. What is MCP?

[00:38:52]We already know and this was introduced by entropic. Okay. The the invention of MCP or the logic or the concept of MCP was brought by entropic and uh then it made a global kind of a thing that anyone can basically have a MCP. So even chat GPT has MCP, Gemini has an MCP anything can have but it's very easy in cloud desktop to set up. So this MCP can answer these questions. So suppose someone new is coming to this repo and he does not know what this GST is all about. So I'm explaining what this MCP can answer. Okay, that is there files in have this uh two files MCP and you need combine two and combine register. What is the prerequisite? You need to have a Python 3.10 or higher. And uh I guess basically if you have a basic idea of python then in through the command prompt you can always check whether my python whether you have python or not.

[00:39:41]So python- version you'll get a python 3.12.10 so it's a compatible thing. So 3.12 is a test and verified because I tested in 3.12. So I wrote that okay 3.12 is tested and it works on 3.12 but it will work on 3.10 or higher. Then it said how to install pip install. These are all very very basic concepts of python. Okay. Even if you take a 2-hour session on Python on any through anybody you can you can understand this whole uh page. It's very simple. So cloud desktop you can download from here claw.ai it available from I guess it's a very basic uh download of uh thing nothing technical in downloading this simple download and you have a windows system you'll get a windows system thing and it will come.

[00:40:23]So whether the entropic servers are bit slow or what uh so download for windows is there you can directly come and download. Okay. So this is there and then yes it is visible. Okay. And then setup instruction. Now this is the main thing.

[00:40:36]So it's a very easy fourstep setup.

[00:40:40]Okay. Very easy fourstep setup. Simply download the MCP file. This you can download and you keep it in any path.

[00:40:46]Okay. Any path. Suppose where is my MCP file. I kept my MCP file inside of this.

[00:40:51]This is my MCP file. Okay. Project Pune online demo. So this is where I kept my uh MCP project MCP file. Okay. What's the second step? After downloading this place a file here and keep your wherever you keep this MCP keep your combined G2 and purchase register in the same file.

[00:41:10]Okay. It's same folder. Wherever you keep it so look I've kept it in a simple here here only. These two are the one which you have to keep it. Okay. And then column requirement. So these are the column name because look every every tally or mark or any BZ or any other software SAP or something will have their own headers name right purchase register format will be different. So what I did I at least said okay look these headers must be the same because in the code you can't have a different headers logic. So I said the name of the supplier must be this again I guess I should zoom if someone not able to look uh sub zoomed it a bit. Uh so legal name of supplier GST of supplier invoice number final total tax IGST CGST SGST and filing period. So these are the mandatory column it must have and it can be in any order first second third any order but yeah these names should match because only then the code will work because in the code if you see I have used these names okay that is why I am saying that these names will require look this in the code these names are used okay so if if you don't want this code will be with you you can change the code if you have GST and if you want to simply delete this you say no I have something other name I'll just keep this so this is up to you you can delete the code and you can make the if you have like if even if you have basic idea of Python you can very well change all the code okay this is raw file that you will get then find your clause desktop config file so this is the a little bit uh the main thing so this excel we can do excel we have been working always so we know this these columns are required and all and then in the claude when you come in the claw look at the top left in the this menu you have this so I have pro version but you do not need a pro version it is there even in the normal versions as well okay in Pro you won't find in in the normal you won't find co-work and code you'll just find chat but in the below you'll find settings and then in the settings you'll just click on here and then we have this developer option and developer we have this edit config okay so when you edit so I have already loaded this that is where it is uh already being shown but in your case it it might not be so edit config will come so come to edit config and it will open the configuration file of the claude So look clude local cache roaming claude. So here is my cache file. This is my JSON file. Okay. So this is a file you have to identify and then you have to make some changes to this file. Okay. So windows sometimes it will be under this app data roaming cloud and sometime if Microsoft store version like there are two versions of how you have installed cloud. If you have done a direct installer like this which I said it will be under app data roaming cloud and if you have done like a Microsoft store thing which in which is in my case I don't know how but yeah in my case this is the thing the local cast roaming cloud cloud desktop will be there while matching the entire cons only GST number invoice number or both if it has difference invoice number then it consider invoice number as well because I have written the logic in the way that GST number and invoice my invoice date all three things should match only then it will consider match Okay, I've taken three things for it.

[00:44:15]It works uh not for 2B or 2A. Look, it just needs this uh column number. Okay, if you keep any 2A, 2B, if you have these column number, if you have this file and uh we just need two file and this column uh names or headings. Okay, that's all which is required uh that is there and okay, so you found this config JSON file. Everyone will like come to this path, you'll find this config file. Now, how to edit it? So JSON if anyone has ever tried to open a JST tob JSON or anything it will it will have this thing called edit in notepad you will have get this option in JSON edit in notepad when you edit in notepad in your case you will get the things only this one preferences and all only this much will come this will not come okay this will not come because this I have added externally so if you have to add you can like simply edit like editing notepad only you can do enter you can type anything you want all these things you can do but don't type anything It's a configuration file. If it gets corrupted, then you have to reinstall cloud. Okay. So be very careful as to when you type because this is exactly the thing which you have to do. You'll get the data from preferences. You have to add this partic. And this is exactly what I've wrote in this uh readme file. So edit a config file. Add the MCP servers block at the very top before the preferences key. So preferences key starts from here. The existing content will remain the same. Don't need to change anything.

[00:45:35]Just this much you need to add. Now what is this? It's basically this python location where is your python stored this python location is there and the mcp location is there that which python file it needs to trigger okay so how to get this again it's a very like a basic concept in python we which we usually cover so like in the command promp where python so you'll get the location of the python which you have but you'll get a single slash over here you have to replace it with a double slash okay this also we usually cover that double slash because It's considered escape character and yeah uh yes you need to log into GitHub account and look GST1 it will work but what you want to like complain gesture one suppose you have two files this is GST one this is sales register so you can have these uh names you keep these names or headings it will work but what I would say if you have a little bit interest in coding or doing something then maybe you can learn and you can write a very brand new code which will compare eway bill and GSTR um GSTR E invoice and sales register and one maybe it can compare all the four files and give you a summary you can write such MCPS by yourself if you there's a little bit of um like a coding little bit you don't need to code the code will be written by the claw or gemini or codeex or cursor or blackbox there are multiple tools to write the code you just need to understand the syntaxes you know so that's one thing so as I said this is how you do the edits once you've done the edit simply save close the file restart the cloud desktop and when you Restart the cloud will automatically see this GST and lookup.

[00:47:09]So how you will see this? You'll go back and when you do the add button the connectors you'll see GST lookup is automatically coming okay automatically it will come if I've done the settings right everything right it should come when this is on only then it can look now so if I turn it off I turn it off and now if I write this particular thing uh if I write this let's see uh what is the GST of Sarasoti Enterprises. I know this is a vendor there. Let's see whether it will able to ask working.

[00:47:51]Look tool result could not be submitted.

[00:47:53]The tools might have expired or the connection interrupted. Refresh the page to it. See the GST and lookup has been disconnected. However, based on the compliance scorecard retrieve in our earlier conference session, it is uh like smart enough to go early but this could not do that. Okay. So, what I have to do? I have to use this connectors and I have to make it switch on. Okay, only then this will be able to use the tool and then when I say to uh I'll say give me a monthly summary of ITC ads per register and 2B it will give me a summary only then okay so because it will use the tool and it can retrive the it loading the tools monthly it is already finding that function what the function now it's loading tools and giving the answers okay a long question by cost tube. I'll I'll cover this in a while. So now look now it is giving me a way of an artifact giving me in terms of this. So it's a good thing. So this is how you using a tool you really enhance the capability of claude. Okay. It'll give me all the month and gap and all is giving. So that's a thing. So okay moving ahead. So you saw how to set up uh this uh GST GitHub link. Thanks Ush for sharing. He has already said this GitHub link in the code in the chat. It is open for everyone. Anyone can come download this MC file, do the readme and you'll know how to come here. If you find any issues then yes you can contact me. I have given my email id as well. So this is around the concept of MCPS and I hope you are able to see the power of an MCP.

[00:49:30]Okay, it's a data stays locally so fast, super fast. You can have your own logics written. You can add as many as logic as you want. I have added six tools. You can add 60 tools as per your everything is free. There's no limit to how many tools you can add. Python the only one criteria is you need to have the basics idea of Python. And for I think teaching Python there are multiple courses, multiple YouTube videos, multiple online tutorials are there. Definitely you can learn in terms of that as well. So okay.

[00:50:00]Okay. GitHub profile, install dependencies, add the MCP server, restart cloud, test with a plain English query. So this was how to configure the MCP we have solved step by step along with the demo and uh yes so basic knowledge required Python beginners beginner knowledge is required very beginning thing how the my function is defined reading of excel basic conditions and all and anyways as I said the a lot of AI are there to write code for you opening a folder as a project integrating terminal editing and saving pi files this is VS code this is required just for this uh like viewing okay it's not nothing to do with coding nothing to do with coding just for that so you can you correctly there's my GitHub connected if any changes I made I can directly uh add to the GitHub thing and all those things so that is the benefit of using this and automation mindset definitely yes I have two videos uh in YouTube if someone wants to have a view they can have a view on how the auto very basics of Python and automation mindset okay so you can definitely have a view of that uh thing as well uh so 3MC use case as I saw audit practice practice a lot of things are there MCP is like there's no limit to the u use of MCP okay I've shown you one use case you can use it for each and everything like there are lot of use case for MCPS you can do it whatever types of thing you want to automate through cloud and yes so yes I have this code that I've already given I have a compliance checklist of DPDP how cloud data policy works uh there's some uh official documents this is not not for you I won't say this I will say this this one DPDP related mistakes structural mistake which every CF reform is doing where I usually talk more about how you are making a big non-compliance by doing the audit and advisory services as well as accounting and any other services under the same name. You are putting your COP at risk and that's what I cover over in this article. So I'll send this article I'll request ma'am to share it in the group and all. So through that you can have an idea of how the structural mistake every CFM is doing and how EY spent 5,000 crores to rectify this mistake and still they couldn't do. I would ask everyone if you can uh search about project Everest like project Everest of EY what it was and how exactly it failed and how it cost 5,000 crores to ambitious consulting division abandoned to split. So why it's why it was trying to split the EI and that is exactly what um we are doing the 600 million failed EI project. It's a very recent uh article thing and definitely you all should have a watch it give an idea of how we are also having us doing the same mistake which uh the big form are doing and how we should be correcting it. So thank you.

[00:52:38]Thanks. That's my time. And uh any query? I'm ready to take I'll take one from Cost. He's already asked when you call data using MCP to cloud. It's being saved with cloud servers as the data is ported through cloud. Instead of this, if you create a streaml application on local machine and run Python, it will give you output. The only glitch is that it's not dynamic and any additional question need to be coded in Python.

[00:52:59]Exactly. Cost, I think you have got it right. that uh uh stream is basically you have your own fixed dashboard like we used to have powerbi dashboard or we used to have u sort of any dashboard in lucer studio or anything now the benefit of clot first of all I talk about the benefit of why clots should be used it's that it's uh it it has a brain okay your streamllet browser or any other thing does not have a brain it will do 100% deterministic thing it cannot like even if I type the name of the vendor wrong if I instead of global I type something else then maybe stream it will show an error we have to click from drop down and all cloud can do a lot of things it has a lot of brain of which function to use that's the first thing why we should use cloud as an MCP second you are saying that data is said with the clot that's true now the the catch here is the entire data is not said whatever the result of that code is that data is said so it the invoice number you can definitely hide the invoice number and that control lies with you I'll just share my screen once uh That's a very genuine concern and definitely in the DPDP act we say this uh as sudo data data anonymization. So what is this called is whenever you are writing this thing now you can have a suppose you write it by yourself or you write through cloud you you tell that whenever you are returning this return only the don't return invoice number return the amount and the count of the invoice number so you have the control of what data you want to send to claude okay so claude will never read your excel files first of all claude will never read your excel file it cannot read it will just read your code in the code you don't have any client data you have just the logics what logics to do and in the logics the file is within your uh same reach file is within your this uh D drive or E drive which you have the uh the Python is reading the file it is much faster in streamllet if you have 10,000 20,000 50,000 rows it will go much slower you can't uh cope up with that slow slowness in streamllet and even in streaml like you can't share it with someone else if you want to say then again it will go to streamlit server which is more dangerous than uh this thing which we are So in that way definitely yes in a first hint side it might look that stream is much better but it's only better when you have to locally use it. If you have to give it to your articles even they can use this MCP then you simply have to just trans the file is already in your I um in your CF forms server already or they can have locally also but you can just give the Python file. Okay. Does this work on branch tally and client with multiple JSON registration? Look I have nothing to do with tally or nothing to do with any other. I just need cloud desktop needs to be there and you need to have Python. Okay. And I work with Excel file. So I'm not going inside your tally first of all. I'm not going insideally or purch or XYZ any ERP you are using. You have to bring it both in Excel files. You have to have a third Python file. That Python file will read and have all those logics. And then you have to need a claude. Okay. So I have nothing to do with whether it will work with tally or not. Whether it work with any other thing or not. It works on claude free also. And one thing uh one thing I'll again say since if someone is planning to buy the claude so please work buy a uh what we say it a work account okay don't buy a consumer account. So what I mean by this is so this is my chat.

[00:56:23]I'm just going a little overboard I guess but just to show for the perspective of uh if they're planning to buy cloud then I'll just show you one thing.

[00:56:38]Okay.

[00:56:46]Huh.

[00:56:49]Uh okay. Okay. All possible ways to be DPP compliant while using cloud. Okay. So it was saying that if you use cloud you are not DPTP compliant. There are multiple ways because the first thing was crossboarder transfer problem. It stores all data in US server and till now India has not certified US as a place where we can keep the data. So that was the first problem and cold legal reg. So first thing was I I was using cloud pro $20 a month. This is a consumer account.

[00:57:17]Here's the exact clear problem. There is no data processing agreement. Okay.

[00:57:21]because it requires a data processing agreement to have with the claude with any server which are using the DPP act requires us to have a DPA with the claude at the consumer account there's no DPA allowed there's no provision of DPA even you go at $20 a month or $200 a month even the max plan if you go that's a consumer account okay so what you have to do rather is the first thing which is said your current 20 month I don't know whether you're able to see this or not you able to see yeah Reject all cookies. Your current $20 pro plan gives you zero legal protection.

[00:57:55]You need an upgrade. Antropic commercial term states that Antropic may not train on models on consumer content from service that apply and this applies to claude for work at team account. Okay.

[00:58:05]So you have to go with claude for work at team account. Not this 20 month of pro plan. The first thing okay if you are using it for your CA form personally you are using that's fine but CAM clients data for work you are using you have to go with work at team standard or premium plan. The clot team starts at $25 to $30 per user per month. This is not much because already paying $20, $5 increase or $10 increase is there. But for a CFO of your size, reasonable upgrade from $20, you can go this there will be no training on your client data contractually guaranteed. It will sign a contract with you. A data processing agreement will be signed by entropic with you which will be legally binding a much stronger data protection act. Okay.

[00:58:45]this signing will be done and then there are multiple other things like how you can do implement data minimization. So it's exactly how the using of MCP thing is there that whenever you have to work on your data use MCP give minimum data to the cloud okay because when you follow this at least you can this is called pseudo minimization at least you can say whenever a data bit something occurs you can say that I implemented MCP and the logics I have written I have guaranteed minimal data now create a client concept framework and there are multiple things okay and uh like in my usually my DBTP sessions I cover all of these that are how these can be built how you can build a client portal through Python by yourself and there's no need of any high big IT team for having all that but definitely yes that's not the topic for today so okay so this was around this and what is the difference between MCP and a plug-in okay if you're asking for the connectors uh maybe someone is asking for the connectors I did cover that because that's a very user friendly thing to have suppose you have this connectors okay I'll I'll go with this cloud only suppose you have this connectors over here and uh you can have so these are the MCPS which are connected. Okay, this is auto auto. This is my GST lookup.

[00:59:52]This is called MCP. Now, if you say connector, so connectors are like we have uh what you say this Google Drive, Gmail, uh Figma, notion all of these.

[01:00:03]So, what I can do if I want to directly send an email to a uh this thing. So, suppose I I connect this. Let me try.

[01:00:10]Okay. So, this is these are the connectors. So, these are called the kind of a plugins you can say.

[01:00:17]So the plugins is different option but just to give you an idea I'm just uh doing with the Gmail thing.

[01:00:24]Um let me add Gmail ones.

[01:00:31]So these are very like easy to use nothing technical. Uh okay there's account mismatch but different look it's nothing but it's just that you uh do a normal sign in and when you do a sign in you can connect. Now I don't want to connect my uh Gmail actually. So you can just connect here and then when your connector is connected you can directly send email. So what you can write you can write okay send this email to this particular email id. It will automatically send using your connector.

[01:00:56]Okay. So that's the simple meaning of connectors. So yes any other questions?

[01:01:00]Let me see.

[01:01:06]Okay.

[01:01:12]Oh yeah, I'm getting questions over here. Uh, plugins. Yes, yes, yes. Okay, take someone asking cloud free or paid.

[01:01:23]Let me ask the questions.

[01:01:26]Huh? So, can the recent be done party wise instead of summary? Yes, Gory, it can be done in any level. It's up to you how you want to do it. You can I had six tools. You can have as many tools as you want. Okay.

[01:01:39]Second was clot free or paid in both it will work. AI will see your data in summary in cloud. Yeah. So the summary data which still see the charts and all that can be a issue with the cloud but again if you have just sending numbers that's not an issue. You're sending the total numbers of the uh sales that the sales number because any personal data of the client should not be saved. Okay.

[01:02:01]When you're sending numbers and all in without sending the client's PAN number or identifiable information that is not an issue with the DPDP act. Okay. Uh and then difference between MCP and plug-in as I said that with the ready to use things the are mostly connectors and all that usually I referred as plug-in and yes that was it from my side I stop saying any questions I'm ready to take I guess I've answered those for the one who were there in the chat great nice so okay thanks informative thanks thanks to glad that you liked it Okay, thank you. Great. So yes, I would say the the era of AI has already started and a lot of era of compliance has also come. So we have to balance efficiency and compliance. Both of things has to go hand in hand. You can't do one at the cost of other. You can't say okay I won't use claude. I will go my manual thing. That's a wrong statement. Also you can't say that I don't care about DPP act. I'll go and use cloud. I will upload the things because I have to be more efficient. So it's not about being efficient or being compliant. It's about being both running both things parallelly. Okay. Uh yes my contacts I am I I'm not sure whether I'm not in the Pune group or not but I guess ma'am can say that I'll share you my uh LinkedIn uh profile or I'll share you my this thing. Okay [email protected].

[01:03:29]Yes. So I've saved him like this and uh yeah I I I'll share a link to my LinkedIn profile in the chat. So ma'am this much is from my side. Uh you can take over.

[01:03:45]Okay I'm sending my LinkedIn connection as well. So we can definitely connect if someone is interested in some uh videos related to automation and all. I have more than 100 plus videos in my YouTube channel efficient corporate. So I'll send that I'll maybe give that link as well. You can have a look. Okay. Yes, that's it from my side.

[01:04:16]Okay. So um take ma'am are you there? Yeah ma'am.

[01:04:31]Okay.

[01:04:42]Okay. M have sent this YD channel link.

[01:04:45]We're going to have a look at there as well. And there are a couple of videos which I would strongly recommend you to look. Let me Okay. Let me show that video. Yeah.

[01:05:19]So hi I had this a few videos which you can have a look is for Python concept you can definitely look at um this your office runs on manual work. This is one the automation mindset wake up CF forms and why automation replacing the manual work. So this is one a very good video you can have a look of how what are the possible journal entry testing app what things you can build uh in Python and all. Okay. So there are many the the entire channel is filled with videos for CFMs only how they and Python can basically interact and they can do this video thing. So yes that's it. Let's see whether Nam has come back. Name are you there?

[01:05:59]I'm not sure of whether she's able to hear us or not but I can take up the queries if anyone has any query.

[01:06:07]I hope you all were able to u learn this MCP thing something new and please implement this because look learning is always there in through there are multiple channels everywhere every day we get something Google launched this Google recently launched two three more spark and all a lot of products so it's not about learning it's more about applying that you actually apply in your practice okay that's where the real value is okay let me call I don't know.

[01:06:52]>> Yes. Yes.

[01:06:54]>> Recording also will be available to look back. Okay.

[01:06:57]>> Yeah. Yeah. Recording will be available.

[01:06:58]I guess the ICA branch is doing the recording. They will definitely make it available to all the members.

[01:07:08]Let me connect to NAM recording. I'm not sure whether it will put in my YouTube channel or not. I'll have to check with NAM but it should be put on the ICI Pune branch YouTube channel if it's there. Yeah, it will pun. Yeah, suppose it should be Pune on the Pune channel.

[01:07:32]I'm not able to connect to Nyam again now. I'm not sure of she might be busy somewhere.

[01:08:10]Okay. Thanks Jan. It's uh I'm happy that you liked it.

[01:08:16]I think I guess we can end the session because I'm not able to connect to ma'am neither directly or through WhatsApp. So I guess yeah brother but I guess thanks a lot. Thank you. Thanks thanks. Thank thanks and bulgi.

[01:08:30]Okay so with this I guess I've said my details and uh we can definitely be in contact. I'm active on LinkedIn a lot.

[01:08:36]So we can connect on LinkedIn or any other platform. Yes. Hoping to hoping for many such interactions and uh getting such teaching opportunities. So yes thanks a lot.

[01:08:48]Can you tell me what automation tools you have made and used for consulting?

[01:08:51]Okay, I guess the uh basic tool which I would say is a document management system and a client portal. If you say if if there are people who are interested then maybe I can show you one client portal. I can have a small demo of how the client portal looks like. So, so it's basically a CA website which I basically make for CA firms and say the small demo.

[01:09:20]So basically why it's made for being DPTP compliant. So suppose it's your uh website it's your uh form website and you have a client login portal like this where your client can come and they can have your email address and password.

[01:09:32]Suppose this is one of your client that comes when they log in they'll get such a dashboard. Okay. And believe me it's not very complex to make it can if someone knows the basic basics of Python if they want to learn Python and all even I have my own Python course which I where I teach Python to CA firms and all. So even if you take that basic course have a basic thing or course you can build all of these uh through use of cloud and all okay there definitely you need a cloud because without that you can't do it because there's a lot of coding but you can do this okay where you have a dashboard they can have a document vault all documents stored in your and this document is connected to your laptop okay it's not stored anywhere in D drive or Google drive or somewhere definitely backup can be taken but it is stored primarily on your D drive and it's getting a thing from there messages you can have the message thing you can have your invoices raised through that invoices is the client can make the payment as well. the razor pay link and all you can connect onboarding can be done through here because look this is the data consent that I was talking about the consent is the primary thing that you collecting of your name as per guidelines as per PMLA act as per DPDP act you can write everything that will store data for these many terms until unless he ticks on all of these he can't continue okay so that is what DPDP act says okay so service request if he wants any new request if he wants file GT for current month he can request he can upload documents from Here only he can submit a request. When request is submitted he can directly upload documents. Whatever documents required he can upload from here.

[01:10:58]Simply upload will be done reupload and then submit the package. Like we have the idea is that the just like ordering food on Zumtro is so easy. The ordering any service to a CFM should be as easy as that. And he should be able to track that okay what my work is going and which uh which stage is going is progress review delivered or not. So that's the idea behind this and it has been made from a idea of how can you be DPB compliant while storing your data locally and all what compliance calendar is there that you can have for them the output history whatever data they have ever given to you or taken from you all these data so this DPP act says that they need to have a button where client can take all his data at once from your system at his own will so there's a download all zip file where all data which he has ever given can be downloaded by him finances dashboard you can give it to him so that you can have a data thing that whichever finances are there this will get updated. The telegram bot this is our document management system. This is a separate product altogether but when someone goes with client portal we also add this as a additional thing. So here he can interact with the telegram bot of your own name and u this through this he can upload the data that through mobile phone he's able to upload data that will data will land directly on your laptop there's no third party telegram server data or all we have taken 100% cognizance that DPT act does not allow storing data in third party server so all the tools which we build we take the first step to give a checky that whether I can defend as a reasonable assurance taken in the DPP act or not okay so that is where the data privacy all of this consent form is there he can give his consent he can make for correction nomination the section 14 of DPDP act says that uh he can nominate a person to exercise on my behalf who will be the nominee so this option had to be given by the CFMS so and every CFM has to have a grievance redal officer of assigned so at least in the appearance you have to stay compliant okay now how much actual that judicial office is working or how much he's doing that's a separate thing but at least you have to stay compliant So with that idea as you were asking that what are the basic products of concern for the CAF forms and apart from that there are a lot of tools available which are already if you watch my YouTube channel those are available you can download it from there.

[01:13:17]Okay so yes can you also tell some AI topics in C that people are working on?

[01:13:22]I guess look for anyone uh Rod G if you want to learn AI the best way to learn is to implement in your office. So what you are doing what problem you are facing any work you find repetitive are that that there's a three R framework which I tell to every learner in my course that one is rule based repetitive and resource draining if any task you are doing is rule based it has a rule like a GST you can say it has a rule purchase register GST invoice number invoice date GST and matching if three things are matching then okay as well not okay so any task which is rule based is repetitive you have to do it multiple times a day multiple times a month or you have to do it once once a year but multiple clients are there multiple clients like annual Gary returns these things are repetitive so rule based repetitive and resource draining resource draining means either more time or my more money is being absorbed in doing that work then if these three are there then it is a 100% automatable through some or some tool if there's no tool available then Python can do it where Python is your last resort The Python is your last resort that when no tools will work in Python and you can automate but it's to cover these three facts rule based repetitive and resource drilling can tally and SAP be linked to clot tally can be linked to clot through MCP and there is MCP already uh that be built by some other developer like I'm building finance MCPS someone else is building tally MCP and here he here also kept it open but now the MCP which has been built Till now it is just having read capability. You can just read data from the tally. It can't write the data.

[01:15:04]But I'll say you there uh I'll ask Nam to add me in your group whichever Pune group you are in. I'll share a very good video of how a tally dashboard was built by someone using a very basic uh python knowledge and he built a very good dashboard that he can just open tally into his uh laptop and uh through the OBBC server 9000 port data will flow to the python using pi obc module and you can do a lot of things you can also upload entries into so that's a very good use case I'll say that video in the group general AI industry what products are people selling CA with AI something like that and in general in the industry like what we're selling people are selling with AI I guess a lot of things and know GST reconcilation or maybe TDS reconation ITR filing all these clause 44 AB reporting so multiple thing that telling bank to tally auto single entry like bulk entry or there are multiple things we're selling but what I always believe is now coding is within reach of a non-coders even non-coder like me I'm a non-coder I'm not from CA I'm not from science background And from being in CA background, I've just been using it to solve my own problems. And through that I made my tool. I wrote my first this GST reconation code I wrote in around May 2021. I wrote this because in my office I had to do GST reconation that time and that time this new rule had come that 10% extra you can take. You can't take the all ITC you can just take 10% of what is matching. So every month we do this and that time 2B was not there. GST 2A was there. So 2A we have to download for 12 months compile all that 2A then compare it with my purchase register and then do the ITC booking and all it was a hectic task for 2 days every month so I built that tool first I used macro but macro was very slow and all so then I learned Python and did it so the best way to learn is or not don't go with any tool I never recommend going anyone with any tool you always see your problem and try building on your own because now coding is very easy earlier it was very hard but now it's very easy invest in some good course and learn uh learn the uh language, learn the thing or see some good videos also. That's not an issue but learn it and implement it. Okay. I'm not sure if I'm there in that group or not. We'll drop you a mail referring to this session. Humble request to share those videos. Yes, sure. Sure, sure, sure. Okay. Uh take Yeah, sure. You can you can share me a mail. I'll send you that video links which I'm talking about.

[01:17:32]Okay. Great. So he came. So I guess it was really nice. It was nice interacting with everyone. It's the first time I'm taking a session for Pune ICI. I've taken for WRC before but yeah Pune was no I've taken for Vikasa before. I guess it was a Vikasa session. So yes thanks thanks everyone who has uh joined and stayed till now and yes it was great having all of you here. I hope you all learned something and please apply. My only request is please apply to your practice that you are able to at least build and if you have any issue I'm always there to solve and always there to solve any of the problems that you might have. Okay. So thank you everyone.

[01:18:11]Thank you for the great uh response and being so interactive during the session.

[01:18:16]So okay. So I guess I'll I'll leave the meeting now and there might be some issue at Namson that's why she's not able to respond. So okay thank you all.

[01:18:24]Thank you for joining. Thank you. Thank you. Okay.