AI Tools

[00:07:41]Uh G Nar audible.

[00:07:47]>> Yes sir. Voice is audible.

[00:07:48]>> Yes sir.

[00:07:59]Hello Jatin sir audible.

[00:08:02]Yes sir. My boy to you.

[00:08:22]Welcome en USA students.

[00:08:50]Department of Education, Vidya Bharti, Institution of Training and Research, online workshop.

[00:09:21]organiz Totally based easy.

[00:10:28]Shika Vijay.

[00:10:47]Name Jalander Mochetra University already systemuces.

[00:11:42]Dr. Gupta professor Dr. Basic students.

[00:12:39]participants Google Dr. artificial intelligence practicing tools, natural language processing, wireless communication, image processing active member Delhi research orientation workshop session.

[00:13:26]So Dr. Jatin, I will pass the session to you.

[00:13:32]Thank you.

[00:13:33]Okay, good evening everyone. First of all, I'd like to thanks to all for joining this session and uh a special thanks to Siksha Makum for taking such initiatives. I also like to mention a special thanks to Dr. Vishal Goyel, Professor Punjabi University.

[00:13:56]Because of Dr. Vishal Goyel, I am able to learn all these things. So, thank you very much sir for joining.

[00:14:06]So coming to the workshop. So let us start with the workshop now.

[00:14:22]I hope this screen is visible to all of you.

[00:14:28]>> Yes, it is visible.

[00:14:29]>> Okay, it is visible. So uh this is just a tit containing one slide because rest of the session we are planning with the hands-on practice. So coming to first to set the base for this workshop. Let us understand what AI is.

[00:14:48]First of all AI is probably a biggest buzz word today.

[00:14:53]Yes or no? AI is one of the buzzword. I feel everyone is behind AI.

[00:14:59]everywhere whenever anyone start to start with this research research paper writing in teaching in school in universities colleges everywhere the goal is how we use AI and ease our work but trust me most of people think when they are using chat GPT one of the most widely used tool and they are just typing prompt on chat GPT for their desire add output they think that we are a AI expert this is one of the myth being expert in chat GPT doesn't make anyone are expert of AI tools because the reason being is most of the LLMs I place charge GPT under the category of LLM and most of the LLMs always suffer from certain hallucinations s that is returning false data also suffers from sharing some fabricated data bias data there are multiple issues so according to me if anyone using charg claw gemini any lm without including any manual effort it make a user handicap trust me we end up with bad results bad research bad projects.

[00:16:30]So what the need is use AI tools but also with the manual and human intervention to verify the return facts.

[00:16:43]This will make you hero instead of making you handicap.

[00:16:47]So basically what is artificial intelligence? Artificial intelligence I can say is a broader idea with the goal that we need to make machines intelligent like human being where the machines can learn from data recognize patterns making predictions and understand the language for a better decision making.

[00:17:13]So basically I can in simple words I say that machine learning learns pattern from data.

[00:17:20]Another word if we go beyond machine learning the another word is deep learning. All of you are already familiar with deep learning. So there is a big difference between machine learning and deep learning. Deep learning uses neural networks which are quite complex than machine learning.

[00:17:40]Also with deep learning we can introduce generative AI.

[00:17:47]Most of you already familiar of generative AI. Generative AI means when we create new content such as text and image or video from the existing content.

[00:18:03]Apart from this you frequently hear words LLM, AI agent, agentic AI.

[00:18:11]Before I go with hands-on practice, it is very important to understand the difference between all these three terminologies.

[00:18:20]What is LLM?

[00:18:24]It is simply a large language model where a large corpus of data is created and a model is trained to work and return the answers based on user queries like you're using chart GPT gemini claude deepseek there are multiple LLMs but moving forward to LLM then there is an introduction of AI agent can anyone tell Tell me what is AI agent?

[00:19:00]Okay, let me complete it. AI agent again is capable and powerful than a LLM to Rajiv chatbot. Yes, we might categorize uh AI agent under chatbot depending upon the various applications. So if I move forward basically an AI agent goes further it can use LLM together with tools not we are using only LLM but with LLM we can use tools it has data memory to perform certain actions.

[00:19:41]So it is quite powerful than using a basic LLM.

[00:19:46]Apart from this, I talked about agentic AI.

[00:19:51]Agentic AI take this further much powerful than a basic AI agent.

[00:19:59]It can define the workflow and plan multiple steps.

[00:20:08]Based on that multiple steps, we can complete our work.

[00:20:16]In a basic example, let's say if we need to plan a visit from Delhi Prince Colony to Mumbai for tomorrow. If we feed this input to a LLM, LLM can simply suggest us you can take this flight, you can take this cab, you can reach to this airport and then you can uh reach to this location, you can book your stay at this location. That's it. This is the output of a LLM.

[00:20:49]LLM can share your planning but execution part is missing.

[00:20:56]Coming to agent AI or AI agent, same prompt to AI agent. What AI agent can do for you? AI agent can book a flight for you because it pre- know your preferences or we can define our preferences. It can book a flight for you. It can book a cab for you from friends colony Delhi to IGI airport Delhi. It can book a cab for you again.

[00:21:24]It can book a cab from Mumbai airport to the stay or hotel in Mumbai. AI agent can execute this workflow.

[00:21:38]Now question is if AI agent is this much powerful then what is the difference between agentic AI?

[00:21:47]The issue is let's say AI agent just book a cab from friends colony to IG airport and due to certain reason that cab is canled in this case AI agent has no capability to again run the workflow.

[00:22:13]The task given to AI agent is still uncertain.

[00:22:19]The completion of that task depends upon the delivery of all the best cases that we fall in best cases for all the events. Coming to Agentic AI, whereas Agentic AI, if for that instance a cab is canled, Aentic AI will ensure it will rebook the cab and ensure you will reach to Mumbai Hotel.

[00:22:51]It can execute the workflow again and again and ensure that task must be completed.

[00:23:01]getting my point?

[00:23:05]Any doubt?

[00:23:10]Because after this we need to go with implementation part.

[00:23:19]So now let us start with our implementation part.

[00:23:25]So I request everyone please open the first tool we are covering that is Google Collab.

[00:23:34]This screen is visible to all of you.

[00:23:35]Google Collab.

[00:23:39]>> Yes sir.

[00:23:40]>> Okay please open Google Collab.

[00:23:45]If anyone has Python version in running in local you can use that but I recommend to use Google Collab. The reason being is it is already bought with Gemini. So it will ease our task.

[00:23:58]Also I request everyone to be on laptop please to make this workshop more sensible and helpful for you.

[00:24:08]Please open Google Collab. Under Google Collab you need to start with a new sheet.

[00:24:23]So this is the screen where you land in Google Collab. It require your Gmail login. Please login using Gmails.

[00:24:33]Then you need to click on new notebook.

[00:24:46]This is our final page where all of you need to land.

[00:24:52]Take 30 sec 30 more seconds so that all of you are on same page.

[00:25:10]Uh this screen is visible to all of you in full mode.

[00:25:18]Yes sir.

[00:25:20]>> Okay.

[00:25:27]I hope all of you already open this Google collab.

[00:25:33]So our first goal is to learn instead of using the simple chat GPT or etc models that how we can simply work on collab okay later on we can also take the help of Gemini but apart from this there are certain learnings which require certain human efforts.

[00:25:56]So let us start with a basic program to understand how a machine learning model work.

[00:26:04]If you are beginner of coding or you have no prior knowledge of coding no issue. This work can again be replicated by AI. But we need to understand what the difference between when we are writing a code and uh ask AI to improve it or we are directly generating a code using AI. What's the difference? It is very important to understand.

[00:26:26]So we start with a very very simple code. If you have a very beginner knowledge of Python, you can go for it.

[00:26:35]If you have no prior experience to use Python or no knowledge of Python, still no issue.

[00:26:42]This workshop enable you to learn how we can write basic codes in Python.

[00:26:50]So for first 20 minute for the next 20 minutes the planning is to start with coding part then we go with the direct AI tools.

[00:27:02]So it will basically set a foundation for all of you that what is the difference between a manual coding and the codes generated by EI. I start typing from skarn.

[00:27:21]Skarn here is basically a Python library which I am directly calling here.

[00:27:29]It is simply a library like in C, C++ all of you already aware of that you import certain libraries like IO stream. So it is exactly the same skarn dot. Let's say we need to call a model.

[00:27:46]All of you are aware of multiple machine learning models. SBM, random, random forest, gradient boost, everything. All of you already know about that. So I'm starting with SVM.

[00:28:03]Always remember from skarn. SBM I'm just working on a single header file. Rest all are the logics.

[00:28:14]SVM always imported like SVC.

[00:28:19]So we are importing SVM with SVC.

[00:28:28]Once we done importing it, let us define certain things. Let's say we are taking a variable X.

[00:28:40]Under X, let me define certain data. Let us take some student data. Let us plan it. So let me take uh one is the student study R.

[00:28:55]Let's say 40 is the student attendance percentage.

[00:29:01]One is student study are 40 is student attendance percentage. Be careful about this screen. If you want I can zoom it now. Better visibility.

[00:29:15]or I zoom it more.

[00:29:18]Okay. Yes.

[00:29:20]>> This is the first data I taken that one is study 40 is the percentage of student attendance. You can take any values as per your wish.

[00:29:34]This is very simple. I'm simply taking a x variable where I'm considering these values.

[00:29:40]Then other value I take is let's say we take study hours to 2 hours and percentage of attendance is 50 another value.

[00:29:55]Now we go with study hours four and improve the attendance to 70 and uh another data let's say we take study hours to five and attendance is very good that is 90%. Now up to this line I hope all of you completed and typed on your screens. You can take any values.

[00:30:22]You do not need to follow the exact values I'm doing for better understanding. You can take any values.

[00:30:32]This is the input data which we need to give to machine that student study R is one attendance is 40%. Based on this data, I need to make a machine. Uh these are just comments. Please do not type it.

[00:30:51]Machine learn that uh student will pass or pay.

[00:31:09]Getting my point? Based on this data what I want that my machine will learn based on my input based on a training learning the patterns it will understand if I ask providing that this is the number of hours of the student study and this is the attendance. Can you tell me whether student will pass or fail?

[00:31:35]So how ML will learn the patterns?

[00:31:39]This is a single input.

[00:31:42]But ML requires labeled data. ML need labeled data which we can say ground truth.

[00:31:59]Ground truth means for this data I need to tell also input the actual label with that like let us label this data. Now this is x. Now we pass another Y to the machine.

[00:32:19]For Y we can say that. Okay.

[00:32:22]Let us take another. Now uh let's say for this fail student study R is one, attendance is 40. The ground truth is fail.

[00:32:38]Okay, we can take this in double quotes because we need to define this is a string type data. So every time you are defining any string type data, please use double quotes in Python.

[00:32:52]Fail for another study are two attendance percentage is 50. Let's categorize into fail category again.

[00:33:03]So student study R is 40 sorry four and attendance is 70 it is a healthy value so we will categorize it as pass another is more healthy value so I again categorize it as pass so this is the ground truth associated with each value the point here to understand is every time we pass any input to any ML model you please associate the label or ground truth with that.

[00:33:39]This is the major problem with all the students that they are simply passing the input passing the non-labelled data and their model end up with a failed learning.

[00:33:54]Clear up to this point?

[00:33:58]any doubt okay now let's use our model so I use my model that model uh SBC I already defined you can see that from SBC I already defined you can directly use SBC and uh we need to fit that SBC to our defined variables One variable is X, another variable is Y.

[00:34:34]I simply pass this to my model directly calling this model. Every time we use any SPM model, we need to use it with the SVC and uh but same doesn't go for other models like random forest. Every model has it different syntax.

[00:34:52]So simply what I need to do is let us print it.

[00:34:55]Our model predict model I already defined you can say it it already suggesting me everything see on the screen whenever you are working on collab it make your work quite easier look at this model.predict Predict then providing any random input by default it completed my code.

[00:35:27]So we need to predict what's the classification of a student fail or pass. A student who study for 3 hours and attendance is 60%. Now let us run it.

[00:35:50]All of you please please execute it.

[00:36:02]Uh all the participants are requested to mute their mic.

[00:36:28]Okay, I manually mute it.

[00:36:31]All the participants please always try to mute their mic whenever they are joining otherwise everyone will be interrupted.

[00:36:40]So you can see the classification is fail. For a student whose study R is three, attendance is 60, it will categorize it to fail.

[00:36:54]It is the prediction but it will not remain same every time. So let's say a student study for 7 hours, attendance is 60 or let's say student study for 3 hours and attendance is 80%.

[00:37:14]I keep study are same improve the attendance. Now let us check again execute it pass. So the thing you have to understand is every time the decision of a ML model will be changed based on to our the based on our input.

[00:37:38]ML will try to understand the patterns.

[00:37:42]So we can say that it is a healthy ML model.

[00:37:48]I changed the statement. This is not a healthy ML model there. Yes, it is ML model. It is doing our task. It is classifying the things. But the data here is very small. Very small. Based on this just chunk of data, the ML model learn bad.

[00:38:09]So the statement everyone of you already heard of that to train any model we require a surplus amount of data is true.

[00:38:21]Adding into that statement with that surplus amount of data we also need to ensure that that data must be of good quality.

[00:38:35]So we are classifying it as this is a classification problem or I can also say that it is a prediction problem.

[00:39:01]Please mention in chat box it is a classification problem or a prediction problem or both are same.

[00:39:08]This is one of the biggest thing all of you need to understand the difference between classification and prediction.

[00:39:14]Please uh type your responses in chat box that it is a classification problem or a prediction problem or both are same.

[00:39:33]classification.

[00:39:35]Can anyone tell me what for prediction law I added this cell uh one of the response I received in chat that uh we need to change this value to 250.

[00:40:02]So it's fail again sir.

[00:40:09]So all of you are understanding very well. Yes, this is a classification problem. This is not a prediction problem. And prediction we need to predict something.

[00:40:20]We are classifying here based on the inputs. We're not predicting anything.

[00:40:29]Let us also check one prediction problem. Then we straightway go with the AI tools usage.

[00:40:39]So all of you done with this classification problem. So can we move to one prediction problem?

[00:40:51]Can we move to prediction problem?

[00:40:54]Any issue in this code?

[00:41:00]If all of you already done with this code then congratulations you have trained your first ML model just in five lines. You have trained one of the good model that is SVM model just in five lines. [snorts] Now trust me if we give a prompt to chat GPT that no uh let us hit live uh write a code to predict student as pass or fail.

[00:41:52]based on some data like study are comma attendance.

[00:42:17]write uh pull python code use SVM model to train the system.

[00:42:38]So let us check the output for this.

[00:42:59]Can you see it all of you? This code is visible to all of you.

[00:43:11]Yes sir.

[00:43:11]>> This is the response of charge GPT which is very much problematic.

[00:43:17]The task which can be completed in five lines unnecessarily it will be complicated by using a LLM.

[00:43:30]This is the reason to do a realtime exercise in collab to make all of you understand that just relying on chat GPT you will end up with the things that a human without the knowledge or a beginner level Python cannot understand.

[00:43:50]We cannot explain this code. It is a very difficult code for a complete good level of programmer who is very much familiar of Python and a experience of five six years he can understand this code.

[00:44:07]This problem again faced by many research scholars, MTech students or any person who is interested in coding.

[00:44:20]This input very frequently comes to me that sir I just code this using charge GPT and uh my co-supervisor or supervisor just scolded me that can you explain this code and I'm not able to do do that a task which can be completed just in five six line and this is the end result for charg if I ignore the data tabs if I ignore these lines Still the length of code is quite big.

[00:44:55]If I ignore this input, still unnecessarily commands, unnecessarily adding of error handling unnecessarily.

[00:45:07]In same way, let us try one more problem.

[00:45:15]A problem of linear regression.

[00:45:19]Linear regression is one of the model which is used for the prediction of the data mostly.

[00:45:27]So let us start again with a very very small code. We start with that >> linear regression.

[00:45:35]>> Go with the linear regression code.

[00:45:39]I request again all the participants please mute yourself or because it is disturbing me I need to mute manually.

[00:45:54]Okay. Now start with me again from skarn already we know that it is a library of python we need to import. You can already see that it is sharing me the autocomplete option but we need to use linear regression now but not SBC to check the difference from skarn dot linear model it is autocomp completing the things import logistic regression I press the tab it auto complete my what I need to understand that which model I need to use logistic regression how we need to import it from skarn library once we type from skarn dot it will autocomplete you do not need to remember much keyword syntaxes if you are very new to python coming to Next let's say again we start with x. So let's say x equals to now we again take the data that is uh inputed study r we can take 1 2 3 4 I'm typing very slowly so that all of you can cope up with me. Please start typing same on your systems so that you can get a better picture and understanding how the logistic regression work and we get a clear picture the difference between a classification and a prediction.

[00:48:11]I simply take these five datas where I'm defining the input of student study R 1 2 3 4 5 Based on this I can provide the actual marks of student.

[00:48:35]A student who is studying for 1 hour let's say is scoring 35%.

[00:48:40]Another scoring let's say 45%, so on let's say 55%.

[00:48:46]studying for 4 hours let's make it 70%, 5 hours let's say make it 85%.

[00:48:54]study R and output of actual marks of the students.

[00:49:03]I define here this data is freely available for us.

[00:49:08]Understand this is a problem. Uh let us say a student is already in seventh semester of uh his professional course.

[00:49:16]Let's say any course be tech whatever and already I have the previous year semester's data based on the input that a student who is studying for 1 hour scored 35 in previous semester for 2 hours get 45% of score. We are simply predicting the things based on the previous data. Now let us again define the model linear regression. We already define please type with me. Oh one error here.

[00:49:59]Here we define logistic regression. We need to change it to linear linear regression.

[00:50:10]model then equals to linear regression.

[00:50:14]You already see that before it it is underlined as red that it is an error because the reason being is we are importing logistic regression not linear regression. As soon as I change it to linear regression it start responding and error is removed.

[00:50:35]Further we need to fit it with the already defined our x and y values x comma y.

[00:50:49]So basically I'm what we are doing here is we are creating a model and training it based on the existing data.

[00:51:00]A very last simple line we need simply need to predict uh let's say print it print model I already defined predict predict what let us completed uh it is suggesting me six so okay we go with six we need to predict that a student studying for 6 hours what he or she can score incomplete output.

[00:51:40]Uh print model dot print You can see based on this prediction a student studying for 6 hours can score up to 95%.

[00:52:29]This is the prediction not any classification.

[00:52:34]Will you understand the difference between classification and prediction?

[00:52:36]Now prediction we are doing for future that if a student study for 6 hours he or she can score 95% but this data is very very small with same example trust me instead of just inputting this data instead of just inputting this data. If we simply change this data to read our CSV to read our CSV file like we can simply need to simple syntax read CSV and file name CSV.

[00:53:32]If we simply do this using simple pandas, we can directly work on multiple problems like prediction of heart diseases.

[00:53:59]There are ample data sets available on kegle that this is the values for multiple users 5,000 10,000 50,000 users for BP this is the sugar value this is their thyroid value etc and etc based on that ground truth is already available that this person is suffering from heart disease yes no or which disease he or she is suffering from with the same linear regression model.

[00:54:31]If we simply change these two lines and make it to read our CSV, model will train on that CSV file and you end up returning up with a good trained model which can predict medicine, heart disease, thyroid disease, Alzheimer disease, Parkinson disease, etc. and etc. And trust me most of the persons who are into the research writing their research papers research papers are still publishing this kind of research a code which we can complete in maximum 7 to eight lines being with predictions if I am ignoring that plotting of that confusion matrices plots of accuracy etc and etc that may include more than five to six line more I take max of 12 lines we can complete this code within 12 lines and go for publication and it is predicting well.

[00:55:36]Once you do that you find a very good accuracy if your data is very much clean.

[00:55:43]So this is the power of machine learning.

[00:55:47]Instead of just relying on chat GPT for your codes, use very very simple methodologies.

[00:55:59]Just two lines in this code. One and two. Rest all our inputs or printing statements with two or three lines max. We are using or we can use any ML model. It may be random for us. It may be SBM. We already seen this is logistic regression for prediction.

[00:56:28]I hope all of you understand this.

[00:56:34]Yes or no? Can I move forward?

[00:56:39]Okay, that is good. Hope all of you understand this to make your work very very on your laptops. Open Excel. All of you please open Excel.

[00:56:59]Let me teach you a last exercise of coding where we simply just need to read a CSV file so that you are in a state that you can directly download a data set and make your Python code to read that CSV file because once you are in a state that you can make your system to read that CSV then only you can train the model. It make more sense. So this is the reason. So please open Excel.

[00:57:27]Let let us take any random data. Uh let's say this is id or uh name class any random letter. So let us populate the rows with sorry again with two six then a sorry let's cap slot it for a better visibility let me zoom it Then A B C A F B plus let's say first please populate the rows with any random data you want. And we need to simply file save as save this worksheet as CSV.

[00:58:51]And we need to save it. Let's say I will save it to my uh music folder.

[00:59:04]I simply save it.

[00:59:11]Now let me explore my music folder.

[00:59:16]Uh it contain book one that is in CSV format. Let's make it student.

[00:59:32]I simply created a CSV file.

[00:59:38]We can simply upload this CSV file. Please click on this folder option.

[00:59:48]Once you click on this folder option, you can upload your CSV file here. This is student data. I just uploaded this CSV file.

[01:00:01]Now you can see my CSV file is visible to me.

[01:00:08]Now let's do a code how we can simply read a CSV file. It simply need a pandas import pandas as PD. This is a standard convention. PD is not necessary. You can import pandas as anything. You can import as P anything you want. But DF is important that is data frame. Then we let us go with pandas dot read CSV where we need to name that file. What's the name of that file? Student data.

[01:00:51]Student data dot CSV.

[01:00:56]Then let us print this.

[01:01:01]DF uh participants is this code is correct or incorrect before I run this code is this code is correct or incorrect it's already showing me a red line I already told you in beginning of typing the first line that I'm importing pandas pd but every time it is not necessary I can directly import pandas directly and use pandas here.

[01:01:34]You can see that red line is omitted.

[01:01:38]If we are importing pandas as anything as any variable name then we need to use that variable name. So let us check we run it.

[01:01:50]You can see my data is displayed here in same way. Now let us move to your kegle.

[01:02:07]Most of you already are familiar of kegle.

[01:02:14]Now let us search a data set.

[01:02:19]heart disease.

[01:02:23]Let us find a very small notebook for the same.

[01:02:30]So this is I need to go with data sets.

[01:02:34]Uh let us check is it in CSV format.

[01:02:39]Yes, it is in CSV format and this is just a 35 KB file. So let me download it.

[01:02:47]Download content as zip.

[01:02:51]Download.

[01:02:54]Let us open it.

[01:03:03]We need to extract all.

[01:03:10]This is the hard data.

[01:03:13]Let me place it in music folder again.

[01:03:20]Let me show you what this data containing.

[01:03:25]So this is our data. You can see uh too much records. 919 records based on heart disease. Yes or no. It is already containing a ground truth. chest pain, resting, cholesterol, fasting, resting, max heart rate, exercise, old peak, ST slope, it contain multiple factors along with age. So based on this data, you can directly go for the predictions.

[01:03:52]Now let us see again.

[01:04:03]We'll go to this folder. We upload our hard data but this data is quite big 919 rows.

[01:04:17]So hard. CSV it displayed me the complete 918 rows containing 12 column. Once you are in this state, your work is very very easier. Simply you again need to recall the model we already trained linear regression for prediction or uh SPM for classification.

[01:04:49]You can go with any model and provide this data and we can train our model.

[01:04:58]This is clear to all of you.

[01:05:02]And again congratulation to all of you.

[01:05:04]Today you train two ML models.

[01:05:10]Two ML models. One is linear regression that is then another is SBA model.

[01:05:19]Both the models are trained by all of you.

[01:05:24]Before that most of the people are just relying on this chat GPT which start sharing you along very difficult to understand quite prone to errors these type of codes which requires a very heavy manual intervention to correct that code.

[01:05:50]you start working on a code or any project using chart GPT it will take ample of time because you end up with error again you saw that error again you pass that output to chart GPT again it will find a solution for you a code which can be converted in just five to 10 lines it create a very much chaos for you so what my initial recommendation is use chat zity that is good for the understanding of the research problem or to formulate anything but do not blindly rely on any LLM. This is not about chipity. I'm not criticizing any specific LLM model but chip is most widely used. This is the reason I'm again and repeatedly naming chart GPT.

[01:06:43]Apart from this, now let us understand the power of Google Collab.

[01:06:55]Again we start with a new worksheet uh worksheet of Google Collab.

[01:07:04]All of you again open Google Collab.

[01:07:07]This is specifically for the target audience who are not into higher education or uh they simply need to do their regular tasks like generation of any question paper, generation of any records, anything you need to do.

[01:07:32]The participants here are mixed of multiple ages from multiple uh different backgrounds. This is the reason the initial part we covered is uh completely for the engineering people and rest of the people can understand the complete process of ML. Now I'm switching this workshop for the meanwhile for the people who are not into the higher education but if they are into higher education and they are not aware of using the a IML tool and you need to ease your daily basis tasks or your departmental tasks then how we can do that.

[01:08:11]Now we come to this point as I already told you that Google collab is by default provide with Gemini 2.5 flash. If I need to change this model I can change it to three.

[01:08:27]Frontier intelligence built for speed.

[01:08:30]It is fast and all around help for if you need Gemini 3.1 Pro you need to upgrade your Gemini.

[01:08:40]So let us type any prompt here for any day-to-day application. Let's say we need to generate any question paper uh MCQbased question paper for students.

[01:08:56]We can do it just on prompt basis. Let us find it out how generate a MCQ based question paper for students.

[01:09:20]You can use this prompt according to your requirement for 10th class student, engineering students, MSE students, PhD students, anything as per your requirement. generate a MCQ based question paper for students let's say on topic or you can directly paste your uh course handout the syllabus mentioned in your course handout here directly you can mention that or you can upload that based on topic machine learning and deep learning It contain 50 questions.

[01:10:17]Two marks each with four options out of which one is correct.

[01:10:37]When I simply typed a prompt here that we need to generate a MCQ based question paper for student. So for better conditions I also type here ensure do not look web for questions and ensure no questions directly taken from any websites because also we need to ensure the consistency and integrity for our question paper that uh our question paper does not directly available on any web sources so that student can directly attempt it.

[01:11:26]Based on this prompt, you can further refine this prompt. Share me uh GUI to generate these questions every time and download as word or PDF.

[01:11:53]So let us find it out what Jiminy can do for us based on the simple prompt I mentioned with me. All of you I think already type this prompt and Jiminy start working on this based on our instructions.

[01:12:10]So you can check it start generating where I need to accept and run. Accept and run means once I accept and run it uh won't prompt you again and again.

[01:12:19]Once I click on accept, it will ask your permission for every step.

[01:12:25]So our task is divided into multiple steps. So this is my authenticated system and I'm a single user and sole user of that system. So I will go with accept and run. So now it start to complete my task based on my instructions. You can say it start coding usage of secure APIs.

[01:12:50]Uh so your geminy once you on Google screen this collab screen this is the icon of geminy if still not visible to you okay participants you can also type all of your queries in chat box I'm regularly on to and looking to this chat works.

[01:13:20]Uh this is the icon of Jiminy. If you're not able to launch it on collab screen, this is icon. I am again highlighting this.

[01:13:29]Once you click on this Jiminy icon, it prompt you with this window and it start to work on the same.

[01:13:42]It is asking me for a key. Uh I will again refine the prompt. I had no key.

[01:13:52]You need to do all yourself.

[01:14:07]It starts to complete my work.

[01:14:15]In the meantime, check details from web then generate.

[01:14:38]If it is asking for the key to everyone then it means the reason being is we are simply using a freeware version of collab. Nowadays the issue is whenever we use a freeware version the functionalities are quite limited to us.

[01:14:57]So it require a key generation. We can directly go for the key generation. It is completely detailed here. The how we can generate our API key. We need to click on this icon left of Google Collab panel. Then we need to add a secret key.

[01:15:12]We need to paste that key to the notebook. Based on that, it will start executing me the prompt according to my instruction.

[01:15:21]This is the secrets.

[01:15:24]So you can add a new secret here or you can go for Gemini key or we can add a new secret name question paper.

[01:15:40]So uh I cannot disclose certain values here due to the visibility. It is an interface with where we I cannot disclose this details but all of you can type on your systems because I'm presenting the screen. I cannot show my secret keys here.

[01:15:58]But you can do the same and pass this key simply to Gemini and it will complete your work.

[01:16:05]Once you start executing it will take around 4 5 minutes or maybe 10 minutes.

[01:16:12]Then it will return you your desired output.

[01:16:23]You can complete that when we go with the QA session. Then we also have time because already it's 5:15. Let me cover one more tool. This is a very important tool. All of you please open N8N.io.

[01:16:42]This tool is very powerful, very important. Please open N8 IO on your systems.

[01:16:55]Once you open this website, click on get started for free.

[01:17:03]Then in two or 3 minutes asking you certain details it will create a account for you where we get the free credits.

[01:17:15]Take 2 minutes and complete your login because this tool is very important.

[01:17:22]This tool is one of the best tool I can recommend you for automation of the processes. How we can automate the things. Here we need to understand that can we automate the processes.

[01:17:38]Consider a case where a supervisor co-supervisor need to evaluate hundred of proposals.

[01:17:48]How we manually evaluate every proposal?

[01:17:52]Can we automate this task? The answer is yes. We can automate this task.

[01:17:58]This is a simple example of automation.

[01:18:01]Even teachers can automate their tasks.

[01:18:04]When a student submitted the assignment, can we automatically evaluate that assignments? Yes, we can.

[01:18:13]Now, let us find it out with the least usage of credits. How we can create any framework using N8N.

[01:18:30]Please open NAN then you land to this screen.

[01:18:39]Once all of you are ready and land to this screen and some of you already completed this please type in chat box so that I'm aware that how many participants are already completed this task and landed to this page.

[01:19:00]Once you click on login start for free, it will ask you certain details.

[01:19:06]Please type certain details. It will ask you for password and a unique username.

[01:19:12]You can keep any username of your choice. If it is available, it will be allocated to you.

[01:19:26]Anyone completed?

[01:19:34]>> Yes sir.

[01:19:35]>> Okay. Uh let me share a basic prompt with you instead of wasting time on just typing this. Okay.

[01:19:49]To consume less credits because we are using a trial version. Once you like this version, you can go for its pro version. anything you can do. I simply need to create a simple research readiness workflow. Research and readiness workflow in that student with the workflow is student need to enter his name. Number of research paper read.

[01:20:12]Then whether a data set is available for that problem yes or no.

[01:20:18]After form submission we need to check the following conditions. What conditions we need to check? If the student has read 10 or more papers and has a data set available, set the status to ready for research. Now you try to understand this these conditions are not with students.

[01:20:40]These are with supervisor and teachers only.

[01:20:44]What condition is in case a student mention 20 our condition is 30. Again we can classify it to not ready for research.

[01:20:56]You can set this condition according to your need and you can also add on multiple conditions here.

[01:21:06]We need to create a complete readiness workflow otherwise set the status to need more preparation.

[01:21:15]Show the final clear status. Keep the workflow very simple. Do not use any external emails, external APIs, database of paid services. I clearly mentioned here in collab we already faced the problem of key.

[01:21:32]So let us find it out. I share this prompt first all of uh with all of you in chat box. Just a second.

[01:21:44]You can copy this prompt from chat box.

[01:21:47]Now I zoomed it further I think better for your clarity visible to all of you or I zoom it more ch I make it to 150. I simply pasted my prompt here.

[01:22:18]It is already forward with the LLM models, agent models already. We can define a user workflow here.

[01:22:32]We can define any kind of workflow in N8N automation. Initially it is asking what kind what do you want to automate?

[01:22:41]So I simply mention here the prompt and let us check it need to work on this prompts.

[01:22:52]Can see it start working on this. Please with me all of you complete this prompt and start executing the prompt.

[01:23:10]Then it start to generate a workflow for you and a very good workflow and trust me the complete workflow is editable for us. You can see ready for researchness form which is submitted by student. Then the decision is ready for research which is classified under two ways true or false.

[01:23:33]Under false set need for preparation show need preparation results. If it is ready, show results. This is a simple workflow.

[01:23:45]Also, we can change this workflow. We can add anything in this workflow as per your need.

[01:23:58]Clear to all of you?

[01:24:00]All of you done up to this step?

[01:24:05]>> Yes sir.

[01:24:07]So this is the automated workflow based on your conditions. Now we need to execute this workflow. You can see this executed my workflow. My name gupa number of research paper I studied.

[01:24:25]Let's say pro is data set available. No submit based on for the no status need more preparation.

[01:24:43]The complete headache of a teacher of a supervisor supervisor is eliminated here.

[01:24:53]You can create any workflow as per your need.

[01:25:00]for completion of anything.

[01:25:04]Let's say any student is working on let's say a project. We can simply mention the project conditions here that is SV model ready? Yes. If SBA model is ready, JUI is ready? Yes. JUI is ready with conditions? No. Go with this backtrack. Backtrack to again your conditions or chapter planning.

[01:25:34]Done with chapter 1? Yes. Done with chapter 2? Yes. Done for chapter 3? Yes.

[01:25:40]Chapter four? No. Chapter 5? No. Chapter 6? No. Or we can say yes. Then based on this a student can evaluate. Are you ready for ST exam? Are you ready for endm exam or any exam? Based on the conditions we can mention topic here. Done with this topic? Yes. Done with this topic?

[01:26:02]Yes.

[01:26:04]Based on that we can directly prepare a very very good framework you can add sticky notes here. This is a focus panel. We can delete any node.

[01:26:24]Delete.

[01:26:27]We can add any node.

[01:26:30]Add what? Let's say if the problem is yes, let let us add a AI here. AI agent which can generate a action plan for the student who already done with the research uh research problem. Can we add external tool? Yes, I start with AI agent.

[01:26:50]And sorry ex not execute step.

[01:27:04]Let us delete this AI agent here.

[01:27:10]Again we add a E AI agent here.

[01:27:20]retry on fail execute once. Okay, done.

[01:27:25]We are done with our AI agent. Based on your research readiness problem, AI helps the student to prepare for the better results, better problem, how he or she can present.

[01:27:41]Instead of directly relying on a LLM model, we are adding all these things to our workflow.

[01:27:53]Any error can directly be fixed.

[01:27:59]clear to all of you.

[01:28:02]You can create your workflows sir. AI agent add simply uh delete any component or if you need to add simply you click on add. Just a second.

[01:28:27]It is executing. Wait for the execution to complete. Then I will show you. Yes, execute this workflow.

[01:28:39]Yes, we are ready with our workflow.

[01:28:43]I simply click on delete.

[01:28:46]Click on add. You can add add AI agent back. You can add action in app data transformation flow anything you can add according to your need. If you need to transform the student data to anything you can add a transformation here. Let's say the problem is every time I see this problem that uh we need to collect certain data from the students.

[01:29:16]It may be regarding their placement data. It may be regarding their academic data, any kind of data. But what we get is a very vague data.

[01:29:27]We manually need to correct everything.

[01:29:30]In case we share a Google sheet with students and ask them to fill their information, parents name etc etc with email id. We see that most of the email ids are incorrectly typed.

[01:29:42]We can add a data transformation step here.

[01:29:46]How a data can be transformed?

[01:29:49]We can simply suggest a AI that based on these conditions data cannot be entered in an incorrect manner. It will transform a data for us in a correct way. This is the way to add AI. The question one participant asked how we can add AI. So this is the AI agent we can ask. Once you click on this AI agent, you can add a AI agent here or uh currently we have no data to set it to mock data and save this and click on toss.

[01:30:23]We added a agent here.

[01:30:27]But this agent only work when you are done with and ready with the complete instruction set in the workflow. You need to add only then if without it let us execute it.

[01:30:45]You can simply again type within Gupta and uh 20 research papers uh data set is available. JS data set is available to me.

[01:31:00]Yes, I'm ready for it.

[01:31:04]We are executing the workflow. In case we need to use this AI agent, set the instructions. That's it. Once you're done with your instructions, then once we are correct with tress or falseness, let again let's say go with any false data.

[01:31:34]Once that false data go to this EI agent, ei a agent start to help the students based on the instructions we feed to that AI agent that helps students in formulating the problem, help student in finding the data set etc and etc. But we need to configure that agent. Once you done with that any student if a student is in false state obviously he or she need our help then again and again a student approach to teacher guide to completely eliminate this step. We can set a AI agent here.

[01:32:14]Done with this.

[01:32:17]Can I move to next tool because we are now getting very short of time. I also want to cover one more tool.

[01:32:27]Now you can practice on this tool.

[01:32:31]Okay, please another tool uh one more tool. First start with chat GPT. One of the most widely tools by all of you. This is the most used tools.

[01:32:47]All of you please go to charg.

[01:32:52]Currently charge GPT is full of multiple options. You can see that charg can do anything for you. Now ch is not uh llm now but the further versioning of chpt introduction of codeax introduction of plug-in make it a complete agentic AI model.

[01:33:14]Now it is depending upon us how we are interacting with chart GPT in a LLM mode in a agent mode or in a agentic way.

[01:33:23]So coming to it see the options are plugins there are many many plugins we can directly use all these plug-in including canva gama for making presentation this is for database superbase flea this is lovable which can directly create apps anyone already familiar with lovable sir okay if you are not familiar with lovable. Please uh write the name of this tool lovable based on any required app by you.

[01:34:07]Trust me you really start loving lovable. It can do the magic for you but it come with a very limited credits.

[01:34:22]Simply on Google ma'am you can type lovable and it will land you to the Levable homepage where you need to login. After login it will provide you I think 5 to 10 credits for a day usage if you are in a freeware plan. Once you are uh completed with your research problem problem, if you like this tool, you can again go for its higher version. But you can see that lovable is one of the agentic AI.

[01:34:48]No manual efforts required. It can create app websites everything in a magical way for you.

[01:34:56]So there are any number of tools. So we can directly uh link our chat GP now with Gmail. It can read, manage our emails, share the responses, teams, everything is under a single umbrella of chart GPT more we find it out sites you haven't made any site create and publish your first site with charged on create site we And let us type any draft word here.

[01:35:43]Website for ML learning. I can say ML learning or website for conversion of text in English Hindi let's say uh regarding the security one of the participant asked the security of our Gmail. Your Gmail is secured completely secured. The reason being is because the chat GPT version you are using it is already linked up with your email and you need to login and uh need to provide access to Chad GPD for the same email otherwise assist will not be granted.

[01:36:45]Apart from this, Chad GPD also launched its further version which can uh stay in your local memory that is codeex.

[01:36:55]Codeex is one of the powerful tool, one of the emerging tool. Any one of you already done with codeex codeex anyone.

[01:37:09]Let me show you the interface of Codex.

[01:37:13]Look.

[01:37:16]Uhoh.

[01:37:36]So apart from this in the meantime when chat GB is working please come to research rabbit.

[01:37:44]I think I missed one or two message. You can again type the message.

[01:37:52]Let me check.

[01:38:00]I hope uh regarding the security point I cleared your doubt amit ma'am any issue I'm not able to understand your message actually uh you have mentioned no no regarding what can you please tell me the context for the same then I'm able to help you and uh Dr. Viml Kumar Sony I'm getting HTML code and site for the flow but this flow is not showing any diagram please suggest uh since sometime after your prompt based on your prompt it take 1 to 2 minutes to design that workflow.

[01:38:42]So this is the reason it never immediately show that workflow once your workflow is ready then only it will be displayed on the screen.

[01:38:51]After this session you can practice on the same. It will surely show you the complete workflow.

[01:39:01]So please come to this research rabbit tool. This is specifically for the participants were very much interested in publications and faced many many rejections from SCA journals and looking for potential candidates to publish in Scopus SEA journals.

[01:39:23]Please go to research rabbit instead of directly relying on the chat GPT for your references. You can see it start working for me. How it is creating the workflow.

[01:39:38]We need to choose. You can see it creating bhasha. It by default start creating me a interface. Bhasha named it.

[01:39:49]We need to just simply click what kind of interface we need. Uh let us select this third one.

[01:39:57]Translate to third. Okay.

[01:40:03]It will create a website for me.

[01:40:07]Creating a website for any of your business nowadays I think is a matter of 10 to 15 minutes.

[01:40:17]There is a time when creating a website is a very tedious task.

[01:40:26]Do we need to manually think of the design? Manually code that how we are adding the data. We need to define complete procedures.

[01:40:38]No simply we need to just based on prompts we can design our website in few minutes.

[01:40:48]I hope uh research rabbit all of you are on the same page once you log into that research rabbit website.

[01:40:58]Research rabbit means we need research rabbit for a better literature review or for better research paper discoveries.

[01:41:08]Now what do you mean by research paper discoveries?

[01:41:12]Many of you faces a problem that DOIs, digital object identifiers returned by any LLM.

[01:41:25]When we ask RGB to add citations and references, it will insert. But when a journal check that most of the DUIs are fabricated and are not found data not exist this is the term return of fabricated data with the hallucination and wrong information which results in major consequences poor research.

[01:41:56]If you fall in a very good case, you are very lucky enough your paper is published. Trust me. Now you can see that most of the journals start to retract the paper.

[01:42:08]Retraction is again a very important issue.

[01:42:15]The participants from the publication background, research background uh know this retraction.

[01:42:24]Retraction here again is one of the big issue we are facing. To avoid that retraction let's say plant leave disease detection uh more refined research sir I am answering regarding that. Okay.

[01:42:52]uh nita I'm answering regarding that that how we can avoid the retraction the retraction major reason for the retraction is the fabricated information which we have added in research papers what kind of fabricated information we are adding let me show you uh plant dis plant le disease detection research paper I can see the research from 2024 research from 2000 etc and etc there are multiple researches are available with So let us consider a paper.

[01:43:26]Okay.

[01:43:29]This is uh one of the paper from IDE.

[01:43:33]So site this copy its DOI.

[01:43:40]I copied its DOI and move to research rabbit.

[01:43:44]Under research rabbit I just pasted the DOI. it starts to search for the similar paper the papers we need in our literature review or to categorize our literature review let's say for plant le uh disease first of all we need to study ML model we simply mention or find a DOI of paper which just contain plant leaf diseases with ML models it will share us only those papers then we again need to categorize the papers of plant leaf disease detection with which are based on deep learning We search for a single paper that single paper is called as a seed paper that is a seed paper.

[01:44:31]Once we go with that seed paper, we can directly search our paper. Is any tool available for editors of any journal of reviewing process as nowadays some mostly student preparing manuscript fully with AI use? Yes sir. Many many tools are available. First of all, if we like you writing the paper from chat GPT.

[01:44:57]Once you done using that chat GPT, delete that chat history.

[01:45:04]Again input that paper to Chad GPT.

[01:45:08]Ask Chad GPU to roast that paper. A paper written by Chad GPT. Chad GPT itself find ample of deficiencies in that research.

[01:45:20]Apart from this what we are doing is we are simply writing a paper using LLM.

[01:45:28]Then we will spin the words using multiple spinner tools. What to avoid? to avoid the plagarism and AI detection which is detected by turnitin.

[01:45:44]Turnitin is only tool which is applicable to all the journals.

[01:45:52]What we are doing is we are simply cheating the turnitin.

[01:45:57]Tenitin can tell you the plagarism AI but cannot tell you the references and DOI.

[01:46:03]If we are going for some lowhanging fruits, the low hanging fruits like the journals from Q4 category most of you already I think received messages that uh publish in this journal faces 20,000 that spam messages all of you received and what you see that after span of one or two month the manuscript will not be indexed in Scopus that is waste of money and your research publication because the reason is if we go for certain lowquality journals they are not interested in checking your references if you go for any spring IT SCM publishers are reputed publishers of SCI answer to your question they do not require any tool to check that but their review process is quite rigorous.

[01:47:05]Their reviews are quite trustable.

[01:47:08]They will thoroughly check your manuscript including your references, including your DOIs.

[01:47:16]If we talk about any tool, there are certain tools available which can check your references. Even if you paste your references to a model like intelligent model like claude or you can put it to any agent AI, it will thoroughly check your references and eliminate those references which does not exist.

[01:47:42]So coming back to this based on this research paper uh almost 1.5k research papers find um we can find out here.

[01:47:54]So let's move it further.

[01:48:00]Search setting. This search setting are only applicable if we are using their upgraded version. But let us check what kind of options. Result count we can limit it to 10 specifically for disease detection. Yes, we need only recent research papers.

[01:48:21]Let's say from 2021 to 2026 I mentioned what kind of category qualile Q1 Q2 okay let us go with Q1 Q2 let us go with a good H index uh yes this is a sensible H index and it must be opensis the reason is always select this option so You can download the paper. Nonopens papers cannot be downloaded.

[01:48:59]So based on this you can select the research papers. You can completely move the positions. You can select your research papers and your selected research papers. We can directly transform these data to a very wonderful literature review or we can directly feed these papers to our LLM model. Apart from this for your better citations I am exporting this five citations. Let us check the five paper I selected. At least this is ensured that all these citations are existing and are true.

[01:49:42]This is a direct citations. We can give it to Latex. Letting we need to do just to remove this rabbit ID and this is a completely ready to use citations for your research paper.

[01:50:01]Getting my point?

[01:50:03]how we ensure the correctness of our citations.

[01:50:14]So there are certain more tools to cover. One is we are left with emergentus.

[01:50:23]Okay. But okay now as we are short of time we plan another session whenever our Sika mahakum will plan for the next session we will take another workshop and uh cover further tools that uh how we can create our own training what we are training on local how we can directly build those okay uh let me stop this training cancel.

[01:50:57]You can see plant leaf new build.

[01:51:07]Let us say its name is plant description. Okay, I'm not adding use temple recipe. You can see we can define a model directly here. We can just train the models within few clicks now and we can decide our own model. We can change llama model gamma with 3.1 billion parameters, 3.4 billion parameters. We can select any desired models and every model strength is given.

[01:51:40]We can change our data set. We can define any data set from hugging phase.

[01:51:44]We simply need to just enter the URL of the data set and that data set directly imported to this workflow. We can design our training configuration. What kind of GPU we need to use.

[01:51:57]The major question arise during training is how we get the GPU.

[01:52:03]Then we can configure our model. Then we can run the training.

[01:52:09]Please note this name E R Datus.ai.

[01:52:14]This is one of the wonderful studio which we can use for the training.

[01:52:20]So for the next rest of the tools we will plan any another session.

[01:52:25]So any question answer you can so any questions you can ask me any doubt you can post or please one by one you can unmute yourself and ask your query.

[01:52:37]After that I will end the session here.

[01:52:45]Yes. Any queries?

[01:52:50]This is aai. Please note it name. E R Tatus.AI.

[01:52:57]Here we not need to import or use every model on our local multiple models which we can use are already here with the billion of parameters. We cannot go beyond billion parameters. Also we cannot train a billion parameter model on our local system or laptop. It require a lot of computational resources might be it's GPU usage goes up to uh more than 48 GB.

[01:53:25]So how we manage to get the 48 GB GPU because computation is very expensive.

[01:53:31]So these kind of platform help the students to generate their results faster and do a better training with low cost. Instead of purchasing that hardware just for a single usage, we can go with our data set here, add our data set here, set our training parameters here, configuration here. Then we can simply just need to train that model.

[01:53:54]Once you've done with your training, you can directly download this model.

[01:54:00]Direct download. After downloading, you can deploy it anywhere.

[01:54:06]Once you start working on this later on later you will understand how the things are working. Okay. Any other question?

[01:54:15]>> Thank you so much sir.

[01:54:17]>> Thank you. Any other question please ask if uh any doubt.

[01:54:29]So today to summarize the things we already done covering with the creation of ML model that is one of the best thing I can say for the students for any specific person that you are training and creating your own ML models within a few line of codes with a better understanding that how the things work. Now also I hope all of you are in a state that at least you can download a CSV data set and train on that.

[01:55:10]At least now all of you are in a state that you can get the results from the CSV file.

[01:55:16]A basic problem which is faced by a researcher during the initial phase of their