Transcript for transcript_1BSgMMHfLjv0KqDcsoXv5rK4QtOG1c0h__1_3420_24d32bdcffe8_5.txt

transcript_1BSgMMHfLjv0KqDcsoXv5rK4QtOG1c0h__1_3420_24d32bdcffe8_5.txt

[00:00:02]  nutrition tracker and project on YouTube transcript summarizer.
[00:00:07]  Also, I have made project on
[00:00:11]  employee's health protection model that I have used and that where I have used machine learning models and pipelines.
[00:00:19]  Okay, so like basically you are familiar with Java or Python?
[00:00:27]  Yes, ma'am.
[00:00:28]  I've done my Java and like in my when I was in school, I'm from ICC board.
[00:00:33]  So I was taught Java from 9th to 12th.
[00:00:37]  So at that time, I was very much familiar with Java.
[00:00:40]  In my college time, I switched to Python because everything was going on in Python.
[00:00:44]  So that's it.
[00:00:46]  Okay, so you're saying that you're comfortable with both Python and Java or only with Python?
[00:00:52]  I'm comfortable in both Python and Java.
[00:00:55]  Like I'm much more comfortable in Python, but yes, Java is also in my portfolio.
[00:01:01]  Yeah, we'll go with some.
[00:01:03]  Okay, basically, first, let me give me a quick intro on this round.
[00:01:07]  So this round, it will be like basic questions on the projects you would have done and related to Java or Python.
[00:01:15]  And any frameworks if you are aware of and then we will be having a DSA round so I mean in this round only we will have a DSA where we will be given like a
[00:01:28]  Mostly two or three problems to you.
[00:01:30]  And we analyze the problem solving.
[00:01:34]  Okay, sure.
[00:01:36]  Okay, then we can start.
[00:01:37]  So you mentioned that nutrition tracker you work on, right?
[00:01:43]  Can you give me like high level what it is and on what area you work on?
[00:01:49]  Yes, I'm sure.
[00:01:50]  So, ma'am, nutrition tracker is basically a web application in which users will be uploading a photo of their meal.
[00:01:57]  And then I have used the Gemini model in which the Gemini will analyze the image and then it will generate a response and tell the user all the nutrition value of the meal.
[00:02:11]  Like it will have given an input prompt so that it will give the response in a structured way.
[00:02:19]  And so that like after giving uploading the photo, it will give the information according like the total calories that carbohydrates, the proteins and the micronutrients and everything regarding to the meat.
[00:02:33]  Okay, so what is the AI tool used for this integration?
[00:02:38]  Yes, Gemini ProVision I have used model.
[00:02:41]  Okay, have you faced any difficulties like while providing the data or like while receiving the data from that Gemini?
[00:02:48]  Yes, sometimes the API calling was failing and that was the major issue that I was facing.
[00:02:54]  And other than that, it was all good.
[00:02:56]  And also there was also a
[00:03:00]  like the data set thing like it was i was not able to get data sets for like improving its security so i manually checked its security like uploaded my meals and checking whether it's correctly giving the responses or not
[00:03:16]  so what is the accuracy the data it is providing
[00:03:20]  any meal let's say i take the picture of the meal and i uploaded that and that meal can have a variety of items in that like so how what is the accuracy rate
[00:03:32]  it was ma'am like it was somewhere between 80 to 85 percent it was it was working fine okay
[00:03:40]  So whenever like whenever I'm uploading any image first basically from the front end, I mean from the web portal, it has to be first collected at the server side.
[00:03:54]  So how that happens?
[00:03:57]  How I mean, I mean, is it through rest API or what is the way of integration for that one?
[00:04:07]  Basically, I used here like the Gemini APIs and Google API used for the uploading of this.
[00:04:13]  So first I upload the photo and then the Gemini API will.
[00:04:22]  Gemini API will call a request to the backend and where the Gemini model is and then the Gemini model will analyze the image and then it will give back the response back to the user.
[00:04:32]  Can you segregate me like what is there on the UI part and what is there on the backend part?
[00:04:37]  So you are saying that the image
[00:04:39]  to upload the image.
[00:04:42]  There is some library on the, I mean, first then we will take care of that image and then it is sending to the server.
[00:04:48]  That is what you are saying, right?
[00:04:51]  So even on the server, like one of the ways like we have to have an endpoint there, an API and user has to, I mean, user has to call that endpoint.
[00:05:03]  Called that uh what do we say like basically can you explain me like what is there on the server like in order to communicate this uh front end with the server there should be some way right
[00:05:20]  Actually, I use the streamlit for the for a front end.
[00:05:23]  So through that it was and through that it was calling to the back end like using the Gemini API.
[00:05:30]  So and I used only the in memory.
[00:05:33]  So I don't I haven't used any kind of database storage.
[00:05:36]  So it was just an in memory.
[00:05:38]  And every time we refresh, the memory will be lost.
[00:05:40]  So I haven't used any such major thing as backend, but just a prototype type of thing.
[00:05:51]  Okay, got it.
[00:05:52]  So you haven't worked on any database integrations as well, right?
[00:05:58]  No, ma'am, I haven't worked on that.
[00:06:00]  So it's only an in-memory one?
[00:06:02]  Yes, ma'am, only in-memory.
[00:06:04]  Okay, so what are the difficulties or what are the disadvantages of using in-memory one?
[00:06:12]  Difficulties of using in memory.
[00:06:14]  Ma'am, first thing is because we can't be, we won't be able to go through again to that.
[00:06:26]  go through that
[00:06:29]  images or the previous history we want we go to the previous history of that we will whenever we reload the site and all the data will be gone and it will be a redundant data redundancy thing and
[00:06:45]  we have to perform repeated calculations
[00:06:50]  So that is why it is like by having an in memory, it is not as optimized as it should be.
[00:06:57]  For longer run.
[00:07:01]  Okay.
[00:07:06]  So how about the speed of response?
[00:07:10]  Like if you are using an in-memory and that memory is going to fill up.
[00:07:16]  So in that case, how it impacts the speed of the performance?
[00:07:21]  I need to ask you, like, let's say a call is taking two milliseconds.
[00:07:27]  If memory is full, how much time it is taking to process your request?
[00:07:40]  I mean, will there be any impact?
[00:07:42]  I'm not asking you the exact time to tell you.
[00:07:45]  There will be an impact definitely because the in-memory is not that much.
[00:07:58]  When the results will take longer time because the RAM will be overloaded with the memory.
[00:08:03]  So that's where it will affect the results afterwards.
[00:08:08]  Okay, have you handled any scalability issues with that or it is just a more kind of project?
[00:08:17]  No, no, ma'am.
[00:08:17]  It was just a normal project.
[00:08:20]  It was not like I haven't worked on its scalability because I said.
[00:08:31]  Can you explain like any other project where you are working on the front end and back end also or on the back end?
[00:08:40]  Ma'am, actually I am not much into like that back end because since my course was artificial intelligence and machine learning, so my basic interest was in machine learning and also in software development.
[00:08:52]  But with regards to front end and back end, there was not much.
[00:08:58]  Like I was not much invested in this front end and back end.
[00:09:01]  Basically, you are like you mean to say like you are mostly working on the AI part.
[00:09:06]  Yes, ma'am.
[00:09:08]  Okay, great.
[00:09:10]  Okay, so.
[00:09:13]  Can you tell me the differences like or like challenges you faced while using or while integrating any AI related libraries or AI related modules into your project?
[00:09:27]  Yes, ma'am.
[00:09:28]  Like.
[00:09:30]  Like the models which I have used in my other projects, like there was Gemini 2.5 flash models and Gemini ProVision models.
[00:09:37]  So if it was like there was also a project in which it was YouTube text transcriber in which the I will put.
[00:09:49]  The video URL of like URL of a video from YouTube and paste it on the prompt side.
[00:09:56]  And from that prompt, the API will
[00:09:59]  that you
[00:10:00]  I've used the YouTube transcript API, which will take the transcript of that video and it will give that transcript to the Gemini and it will analyze all those things, all the transcript and give us the summary of all the video content.
[00:10:13]  And it will help the students or any person who is not able to see long format content.
[00:10:19]  Like if he or she has less amount of time to watch.
[00:10:23]  So it will be easier for him to summarize the whole video within text in text format.
[00:10:28]  So that was the idea.
[00:10:29]  And like handling the challenges in this is like sometimes the Gemini 2.5 flash model was
[00:10:41]  like inconsistent, like it was not giving the.
[00:10:48]  The best responses like it can generate.
[00:10:51]  Also, there was latency and sometimes there was inconsistency and sometimes even it was hallucinating.
[00:10:57]  In my nutrition trigger project, it was sometimes it was hallucinating a little before I make it more accurate.
[00:11:09]  Okay.
[00:11:10]  So, as you mentioned, like a YouTube content, like a
[00:11:17]  YouTube transcription.
[00:11:18]  Transcription.
[00:11:19]  So using, let's say, generally for everything, the limit will be there, like up to this much limit, like 10 MB, 20 MB, 30 MB.
[00:11:30]  This kind of limit will be there where it can process with a super speed.
[00:11:35]  After that, there will be some issues.
[00:11:37]  So let's say if I give you like maybe 500 MB video, how the data will process to the AI?
[00:11:50]  Can you repeat the question?
[00:11:52]  See, my question is like, in order to get the transcription from the video, first we have to give this video as a source to something, right?
[00:12:06]  Let's say you assume that you are having one AI model or something to that model, you have to dump this video and AI will process that video and it will give you the output.
[00:12:17]  Okay, so in this case, like, we can go with a small video.
[00:12:23]  If it is a 1MB video, it can process quickly.
[00:12:26]  If it is a 20MB video, it might take some time.
[00:12:30]  So like I'm asking if it is a bulk data, I mean, the video size is large, then how the processing is happened?
[00:12:37]  Like you have to dump this data to the AI model or something, whatever.
[00:12:43]  You have to give this as a source data, right?
[00:12:46]  This video has a source data to something else.
[00:12:49]  So how that happened for a large size files?
[00:12:57]  Okay, I think the model will break it in break the video in chunks and it will upload it to the Gemini model and
[00:13:08]  to that one by one each chunk will be analyzed and it will generate the output.
[00:13:14]  And it will take a long time, of course, because it will as a batch processing type of thing.
[00:13:19]  So, yes.
[00:13:21]  Yeah.
[00:13:21]  So like how this somewhere my I didn't get the answer yet now because my question is like forget about the model or something, whatever you are integrating to that, you have to pass the data in order to pass the data.
[00:13:37]  First, you have to have the data like maybe you are asking user to input video.
[00:13:42]  And user input the video.
[00:13:43]  So after inputting the video, this video has to be passed to somewhere even before starting the processing.
[00:13:51]  It has to be first sent to the server or it has to be sent to the model.
[00:13:56]  So while sending that, how the behavior is if it is a large size file?
[00:14:03]  Okay, let me think.
[00:14:24]  For the uploading part, as I said, YouTube transcript API, which was sending the record that much.
[00:14:29]  What I know is that the API which I've used, like the YouTube transcript API, which is the transcript from the before like sending into chunks.
[00:14:38]  So that is what I have used here in sending it to the model.
[00:14:44]  Okay, okay, no worries.
[00:14:48]  Can you differentiate between artificial intelligence and the machine learning?
[00:14:56]  Yes, ma'am.
[00:14:56]  Sure.
[00:14:57]  So artificial intelligence is like machine learning is a subset of artificial intelligence and in which machine a machine will learn on how to
[00:15:12]  respond and machine will learn automatically like itself, like by training on data sets.
[00:15:19]  Or maybe not from data sets because there are three types of machine learning like supervised and supervised and semi-supervised and artificial intelligence is
[00:15:31]  artificial intelligence is a model which will generate artificial intelligence is that uses that generates responses from inputs and
[00:15:45]  You're saying ML also same thing, right?
[00:15:47]  It will also send that response from the...
[00:15:49]  Yeah, you know, decision making.
[00:15:51]  And while ML is pattern recognition.
[00:16:00]  Okay, what do you mean by supervised learning?
[00:16:05]  Supervised machine learning is a machine learning model which is trained on a data set, like on a annotated data set.
[00:16:15]  It will be trained on a specified data set and upon that the
[00:16:22]  the machine model will be trained itself on the training data.
[00:16:26]  Okay, just as an example, if we train the MI on this ML on a specific data, maybe by giving these food images.
[00:16:39]  Okay, and suddenly if I upload a different image, maybe some scenery or some vehicles images, how it will react?
[00:16:50]  Okay.
[00:16:56]  The question is like if like I have trained the data on the food images and I have uploaded another photo like scenery or something.
[00:17:03]  Yes.
[00:17:08]  It will generate a response that might be really close to it.
[00:17:12]  Like close to the.
[00:17:16]  like similar search for the similarities between the trained items and the uploaded item okay but here in this case it is uh not at all syncing right the food and the scenery if i'm saying
[00:17:28]  it can be like a completely different um maybe maybe
[00:17:35]  It may generate a response like the uploaded image is like not related to the
[00:17:43]  the train data set and it will then it will upon analyzing the scenery it will generate some text upon like this is the these are the food items that are often
[00:17:57]  use or often okay
[00:18:04]  Could you differentiate between the bias and variance?
[00:18:08]  Bias and variance.
[00:18:14]  Yes, man.
[00:18:15]  So bias is...
[00:18:21]  Bias is like errors due to the wrong exceptions.
[00:18:26]  That is bias.
[00:18:30]  And like error and sensitivity that is comes in bias while in variance it is
[00:18:54]  Marian says...
[00:19:05]  Like bias is when bias bias occurs when the model is like very simple while variance is when the model is complex.
[00:19:15]  So that is thing.
[00:19:18]  Okay.
[00:19:20]  Can you explain what does decision tree means?
[00:19:26]  Decision tree.
[00:19:27]  Yes.
[00:19:28]  Yes.
[00:19:28]  So in decision tree, they're like, it is split into like different
[00:19:40]  models and each model will like each set will be divided into training and data set, training and testing set.
[00:19:48]  And then all the models will be taken into consideration and the best model which will give the best result with accuracy.
[00:20:00]  And that will be the finalized output.
[00:20:02]  That is what happens in decision tree.
[00:20:06]  I get a supervising machine learning algorithm.
[00:20:12]  So what are the three factors considered while taking this decision tree?
[00:20:18]  So like the speed or it will be the accuracy.
[00:20:26]  Um,
[00:20:36]  There will be a speed and
[00:20:46]  Overriding.
[00:20:48]  Oh, sorry.
[00:20:53]  Overfitting
[00:20:58]  of the data.
[00:21:02]  I didn't get you.
[00:21:07]  I'm like overbidding as in
[00:21:18]  Like ma'am, in this case, we will take like, we'll choose speed over accuracy, like, and we have to generate more much more.
[00:21:30]  Like splitting of the data, we have to maintain that.
[00:21:34]  So that is one that is one factor.
[00:21:37]  And
[00:21:43]  So if you're depending on the speed, the results can be a different, right?
[00:21:48]  Like one model can give entirely a different result, but in a quick time.
[00:21:53]  Yes, ma'am.
[00:21:54]  Like it will definitely affect the results.
[00:22:00]  For accuracy, I will use random forest.
[00:22:10]  What do you mean by random forest?
[00:22:12]  Yes, so random forest is like is a algorithm in which it works on it like it has a.
[00:22:26]  It uses multiple trees and it uses random sampling and uses multiple trees.
[00:22:37]  And other than that, it works on the principle of like it works on betterment of this.
[00:22:46]  Like first it will.
[00:22:49]  Gather the major error in the first model and then it will work upon that error and then then get to another result which
[00:23:00]  with less error and like it works on basically minimizing that error in which the model is facing so and and it will and it will work until the error is no more
[00:23:14]  Okay.
[00:23:17]  Okay, cool.
[00:23:18]  We'll go with, you mentioned that you're aware of the Java as well, right?
[00:23:24]  Let's have some questions on that Java then.
[00:23:27]  What do you mean by class and object?
[00:23:31]  Class and object.
[00:23:32]  Ma'am, object is...
[00:23:36]  Object is an entity within a class and class is blueprint or
[00:23:46]  blueprint of properties of the objects of multiple functions methods that is
[00:23:55]  and object is an instance of a class.
[00:23:58]  So how the constructor relates to the object?
[00:24:05]  Constructor is basically a default object that is in Java.
[00:24:27]  Is it necessary to have constructor for all the classes?
[00:24:30]  No, no, ma'am.
[00:24:31]  It's not necessary for the constructor.
[00:24:34]  It is already like made by the, it is already made by default.
[00:24:43]  How can it be a default?
[00:24:45]  What if I...
[00:24:49]  Or tell me this like what if I make the constructor as a private one?
[00:25:03]  Make the constructor as a...
[00:25:05]  Constructor as a...
[00:25:08]  Private.
[00:25:09]  Private.
[00:25:15]  If it is private, then objects of that class will not be accessible.
[00:25:27]  You think that?
[00:25:33]  Okay, you mean to say that no, we can't create any objects with that or we can't access the object?
[00:25:38]  Only access, like we won't be able to access using the private keyword.
[00:25:44]  What do you mean by that?
[00:25:45]  Like, let's say I have a class, maybe class vehicle is there and under that I created a constructor with a private modifier.
[00:25:53]  So can I be able to create an instance for the vehicle?
[00:26:00]  No, ma'am.
[00:26:00]  Outside that class, you won't be able to call it from any other side.
[00:26:05]  Like if like there is one thing that is function calling and we call other functions.
[00:26:10]  So and we can call those function because those functions are public.
[00:26:14]  But in this class, since it is a private class, no other function or class will be able to call the.
[00:26:21]  Variables or any other thing which is any other thing which are inside this class.
[00:26:27]  So I don't want to call any variables or anything.
[00:26:30]  I just want to create an object for that vehicle.
[00:26:33]  Is that possible?
[00:26:35]  Yes, that is possible, I guess.
[00:26:39]  Only in the same constructor, the object can be created.
[00:26:47]  Okay.
[00:26:50]  Okay, fine.
[00:26:50]  We'll go with some DSA questions.
[00:26:53]  Just give me a minute.
[00:26:54]  I'll share you the document.
[00:26:56]  Okay, I'm sure.
[00:28:10]  Okay, here in this chat, I'm sharing you the Google document link.
[00:28:14]  You just open it.
[00:28:15]  I will paste the question there.
[00:28:31]  Can you start sharing your screen?
[00:28:33]  Yes, I'm sharing.
[00:28:39]  There is nothing in this document, by the way.
[00:28:41]  Yeah, yeah.
[00:28:42]  I'm just adding the question.
[00:28:44]  Okay.
[00:28:44]  The question you can...
[00:28:46]  Yes, yes.
[00:28:48]  Please share your screen and you can work on it.
[00:28:51]  Sure.
[00:29:10]  Okay.
[00:29:16]  So here is your first equation where you are given with a string and your target is to find the longest substring, which is a duplicate one.
[00:29:26]  I mean, it has it should have more than one occurrence.
[00:29:31]  Okay, I'm just stringing together all the limited strings.
[00:29:39]  on any duplicated substring that has the longest possible length.
[00:29:53]  I have to do it in Python or Java?
[00:29:56]  Anything, it's up to you.
[00:29:58]  You can open it.
[00:30:00]  the online editing so you can do that yes sure i'll be doing it in python okay
[00:30:15]  Like, can we write, can we, can I first write it on paper, like for myself?
[00:30:21]  You can use that Word document so that I will also know what you are thinking, right?
[00:30:26]  So you can, whatever draft work you want to do, you can do that on that Word document.
[00:30:32]  It is allowed for editing.
[00:30:35]  Okay, I'm sure.
[00:30:42]  I mean, this thing, there is no word actually.
[00:30:47]  What?
[00:30:49]  There is no word in this.
[00:30:51]  Actually, this is my friend's laptop.
[00:30:53]  So that's why.
[00:30:54]  No, I'm saying like this word document is there, right?
[00:30:57]  Oh, in this one.
[00:31:03]  Or requesting a direct access.
[00:31:05]  I think I already provided you the permission.
[00:31:39]  Okay, so the idea is that we are at least
[00:31:43]  Yeah, updated now you can able to update now, edit now.
[00:31:48]  Yes, now I have access.
[00:32:09]  First I'll create a function and
[00:32:35]  And then I'll take the length of the string.
[00:32:41]  Espero que mande.
[00:33:27]  I can do this also.
[00:33:28]  Yes.
[00:33:28]  First, I will convert this like these characters into numeric values like for hashing purpose.
[00:33:38]  Okay.
[00:33:39]  Nice, ma'am.
[00:34:22]  And then we'll do a piece.
[00:34:30]  And then I'll create a search function.
[00:34:34]  Search function.
[00:34:44]  Search me.
[00:34:52]  Take edge as 0, rolling hash, and
[00:35:07]  I range the command.
[00:35:59]  This scene will be in hash set.
[00:40:12]  So what do you think is your time complexity for this program you're writing?
[00:40:21]  Time complexity is O and log n.
[00:40:28]  Because we are using multiple loops and also using this sliding window.
[00:40:34]  So it will be n log n time complexity.
[00:40:38]  Okay.
[00:45:02]  Okay, can you try to run it in the compiler?
[00:45:07]  Yes, ma'am.
[00:48:04]  Siddharth, just leave it.
[00:48:05]  We'll go with the other problem.
[00:48:07]  We are running out of time.
[00:48:09]  Okay, I'm sure.
[00:48:13]  to
[00:48:22]  Okay, I'll be giving the other problem.
[00:48:32]  Okay, I have added it in the same Word document.
[00:48:36]  Okay.
[00:48:37]  I'm going to understand the logic of the previous one.
[00:48:41]  Sorry, I didn't get you.
[00:48:43]  Did you understand the logic of the previous question?
[00:48:45]  Yes, yes, yes.
[00:48:47]  I got that, so that's why I'm...
[00:48:50]  Not allowing the time to work it on the compiler.
[00:48:53]  So you can make that.
[00:48:55]  So for this second problem, the thing is like you are given with numbers.
[00:49:00]  So and your target is to remove the K digits from the given number.
[00:49:05]  So that the remaining number should be the smallest possible value.
[00:49:14]  and an integer k return the smallest possible integer after removing periods.
[00:51:37]  I'm clearly learning.
[00:52:34]  One decrease this.
[00:52:40]  Checked.
[00:54:32]  Yes, ma'am.
[00:54:35]  I've used the idea.
[00:54:37]  Hello.
[00:54:38]  Yeah.
[00:54:40]  So the idea is like first remove, I'm removing the larger left digits and like to minimize the value.
[00:54:48]  This is what I'm doing.
[00:54:50]  So in this function, first I have like initialize the stack and then I'll add this to
[00:54:59]  iterate through each digit.
[00:55:02]  And then in this.
[00:55:07]  If the previous digit is larger, then I'm just removing it to minimize the number and just using the stack.pop to remove larger digit
[00:55:17]  and also decreasing the removal count when I'm using the k minus equal to one.
[00:55:23]  So what is the time complexity here?
[00:55:26]  Time complexity is, ma'am,
[00:55:31]  Oh, and same set as just one loop.
[00:55:35]  So how about the time complexity?
[00:55:38]  Yes, so you did it is pushed once and popped at most once.
[00:55:42]  So that is why.
[00:55:44]  Linear traversal is there.
[00:55:46]  So that's why the own time complexity.
[00:55:50]  Okay, what you are using?
[00:55:54]  I mean, the size of the stack, it will also be matter, right?
[00:55:57]  So, and what is the reason to choose a stack?
[00:56:02]  Because stack is like in stack as the leaf property.
[00:56:07]  So in that in this, it will be easier to remove and remove the digits if they are if it is larger.
[00:56:19]  Okay, fine then.
[00:56:23]  I'm done from my side.
[00:56:26]  Okay.
[00:56:29]  Anything for me, ma'am?
[00:56:31]  Any feedback?
[00:56:33]  Anything?
[00:56:35]  Nothing much.
[00:56:37]  Our HR will contact back to you.
[00:56:40]  Okay, ma'am.
[00:56:41]  Thank you so much.
[00:56:42]  Thank you.
[00:56:43]  Have a good day, ma'am.
[00:56:44]  Thank you.