{
  "video_id": "fx0GGAANIus",
  "channel_slug": "statquest",
  "channel_handle": "statquest",
  "title": "Human Stories in AI: Khushi Jain",
  "duration_seconds": 1633.0,
  "url": "https://www.youtube.com/watch?v=fx0GGAANIus",
  "upload_date": "",
  "transcript": "hello I'm Josh starmer and welcome to\nhuman stories and AI with stack Quest\nand lightning AI in this series we'll\nhear about the career journeys of\npassionate AI experts from their humble\nbeginnings to conquered challenges will\nbe inspired by the realworld experiences\nof professionals thriving in the ever\nevolving AI\nlandscape human stories and AI is\nbrought to you by lightning AI code\ntogether prototyp type train and deploy\nAI web apps all from your browser with\nzero setup personally I love lightning\nAI because it makes it super easy to use\nand learn from the stat Quest coding\ntutorials just go to the web page click\non the Run button and Bam you get code\nthat you can play with without\ndownloading anything or installing any\npackages today we have special guest\ncushy Jane who works in data analytics\ndevelopment M at John Deere she's also\nin the master's program at the\nUniversity of Illinois Urbana Champagne\nworking on a master's in computer\nscience data science having recently\ngraduated with her Bachelors kushy\nparticipated in the data science club\nand also completed several internships\nat John Deere so without further Ado\nkushy can you tell us about your journey\nto where you are right now at John Deere\nand the University of Illinois how did\nthis all start absolutely so um I guess\na fun fact um my whole life I I was\nnever that student who's like I'm going\nto be code you know from middle school\nand I love Tech in fact I was very like\noh I'm terrible at Tech and I defined it\nbased on things like can I just like you\nknow do computers what people often\ndon't realize there's a lot to computers\nlike there's the hardware side there's\nsoftware side of course I didn't know\nthat as a student and I was very much\ninto science I was that you know classic\nScience Olympiad geek um you know and I\njust loved all kinds of Sciences and\nthen my junior year of high school my\ndad said why don't you try an a computer\nscience class I really think you'll\nenjoy it and I I didn't want to\ninitially but I was like fine Dad I'll\ntry it um and I ended up feeling a very\nI don't know crazy kind of a rush in\nthat class I just uh I mean I'm not to\nsay that I'm very instant gratification\nperson but I just loved when code worked\nlike it made me feel like intensely\nsatisfied and somehow that was enough of\nof a feeling to just make me want to\npursue computer science or something\nwith coding um and so yeah that was the\ninitial start of it but I ended up you\nknow really changing pth so I didn't get\ninto computer science at uvi\nunfortunately um so I ended up taking\ninformation signs okay um not being sure\nabout it initially but I ended up loving\nit and I I truly believe that everything\nhappens for a reason um and information\nscience is that really balance where you\ndo get to focus on the technical side of\nthings but also the human Centric side\nand how do you make things customer\noriented and how do you present yourself\nhow do you present your data how do you\npresent your technology um and that's\nhow I kind of got into data science um\nand not just learning you know how to\njust code but how to apply your\nknowledge and present in a way that\nappeals to different kinds of audiences\nso it ended up working out really well\nbut I hope long story short that answers\nto some degree how I got here I\nabsolutely love it I mean I'll be honest\nI I have a similar sort of Love of\ncoding where uh to me it's like solving\ncertain puzzles like little you know\nit's like doing a Sudoku or doing a\ncrossroad puzzle you know when it works\nit you just feel fantastic about it and\nto be honest for me even when it doesn't\nwork I almost always learn something\nfrom that and and I also just like love\nthat process of learning so even even\neven a failure is still kind of like\nmildly a success for me um and I also\nlove that you like uh sort of\nthe the data science side of it to me\nthat's always been super important\npersonally uh when I first got my a job\nuh after my PhD I got a job in a\nbiological genetics lab and I knew very\nlittle about those things they were all\ndoing experiments I was just the numbers\nguy but I loved trying to communicate to\nthem and trying to relate sort of the\ndata analysis that I was doing with them\nuh so I'm in in some ways I feel like we\nmight be kindered Spirits in\nthat kindred spirit with Josh that's so\ncool I don't know people know but I a\nhuge huge fan of stat Quest like stat\nQuest got me through a lot of Concepts\nlike radiant descent so um huge fan so\nit's amazing that you're saying\nsomething like this it feels so special\nwell I mean I mean I guess you get the\nidea right the all the whole idea of of\nwhat I love is is I love the\ncommunication aspect uh and I love that\nit's part of a bigger picture and it's\nnot just sort of like coding for coding\nsake you know it's it's about Community\nit's about solving big problems and\nthat's what I loved about being part of\nScience and I and and you being kind of\na also a science nerd like me um you\nknow it gets us it gets us in it gets us\nto be a part of that but in a in a in a\nway that that for me I've discovered a\nlot of people are scared of a lot of\npeople are scared of the analytics and\nso if we can do it for them we're really\ndoing a great favor or a service or just\nreally being helpful and and I just and\nI like being\nhelpful absolutely yeah I completely\nagree with you on that and I think uh\nthat's also part of why um we run the\nIllinois data science club here at uvi\nbut we believe that every field can\napply data science to op optimize\ndecision making so I completely agreed\ncan be a very interdisciplinary field I\nlove it would would you be willing to\ntell us about the Illinois data science\nclub yeah sure so um our very first\nalthough I did not found it um we kind\nof started our first semester together\nwith the pH with the founder my my\nroommate Ria sha who's also a student\nhere at uvi with the same major um and\nour thought was you know we wanted to\nfind a community that was not doing some\nkind of like hardcore hackathon or\nhardcore you know Tech Consulting where\nyou have a client and a deliverable and\nthere's just a lot of pressure wanted to\nmake it a little more I guess for lack\nof better terms a lowkey okay where\npeople of different you know levels of\nexperience can come to a safe community\nand learn some data science um and yeah\nwe're not expecting that people will be\nexperts at the end of this there's\ndifferent levels of expertise of course\nbut they might have learned how to use\ndata in a useful way and in a subject of\ntheir interest we didn't want to force a\nproject or how they want to do something\nwe just want to create some sort of\nguidelines and guide them through a\nproject of their choice and they get to\ndo a showcase and uh present their idea\num and yeah the key I key thing behind\nour Mo uh motto was that uh we don't\njust do data for data we do it to solve\na bigger problem whether it's business\nor societal so that is kind of like the\nuh key underlying theme of our club but\nyeah that's a little bit about us I love\nthat I love that motto too you don't\njust do data for data you do data to\nsolve bigger more important problems I I\nI I feel like if there's one takeaway\nfrom this podcast episode it's right\nthere that's a nugget of Awesomeness I\nlove it um well can you tell us a little\nbit what what you're doing now I I\nunderstand you're a a senior about to\ngraduate uh yeah sure so so now well\nyeah of course this is my last semester\nI'm graduating and if all things go well\nI plan on pursuing my masters in data\nscience here at uiu it's called The\nmcsds Masters in computer science data\nscience and I want to work full-time\nwhile I do it because it's online um and\nget all the experience that I can that\nis the plan so far I love that plan uh\nyou also had a summer internship at John\nDeere can you tell us a little bit about\nthat uh what you did how did you even\nget the internship to begin with um any\ndetails I would love to hear all about\nit yeah so uh I'm very thankful to John\nDeere because they've given me numerous\namazing opportunities not just one\ninternship I interned there twice and I\neven did part-time there um yeah so how\nI got my first internship in soft or\nactually end of sophomore here um well I\ndon't know if I even had a good resume\nor not but they liked all the python\nstuff I had in there that's what they\ntold me um I think all credit goes to\none main class and it's kind of funny\nit's not my hardest class it wasn't like\ndata structur or something it was a very\nbasic data Discovery class and they\ntaught you the basics of python and how\nto apply in a statistical sort of way um\nfor those who are from uvi shout out to\nthe class stat 207\num yeah it's called data science\nDiscovery and yeah I basically learned\nlike the absolute basics of you know how\nto do like SK skarn packages and what to\ndo with your data um how to do feature\nengineer to some degree a bit of feature\nengineering um that whole Pipeline and\nthat I guess was enough knowledge to get\nme my first internship it's fantastic um\nI love all that stuff too that to be\nhonest that was also my Gateway into\nsort of data science in the python realm\nwas was I saw that in in scikit learn\nthey had um all these machine learning\nmodels\nand what I thought was super cool about\nit is once you got your data which took\nforever but once you got it you could\nthen just try a done a ton of different\nmodels on it without like having to do a\nwhole lot more work um and so I love\nthat well that sounds fantastic so you\ngot these internships which are which\nare great can you tell us a little bit\nwhat you did during those internships oh\nyeah I forgot to answer that sorry it's\nokay um yeah so my first internship was\na very classic analytics one felt um we\ndid like customer call sentiment\nanalysis and like you know grouping\ncalls into red yellow or green based on\nyou know how they're feeling um my\nbiggest takeaway from that internship\nwas Data I that's when I truly realized\ndata is what matters the most we tried a\nbunch of you know ml models and we came\nto the conclusion that you know this is\nprobably not going to work out so well\nas of now because um the audio\ntranscriptions the quality of those were\nnot so great um and\nhumans you know on their own were not\nlike people at the company were not sure\nhow to you know classify calls because\nit was not clear so what better is a\nmachine learning model going to do and I\nlearned that ml is not magic um if a\nhuman can't do it at all or not so great\nthere's a high chance an ml model won't\ndo it that well either but it was a\ngreat learning process and um yeah I\nlearned what all it might take to make a\ngood model um all the stuff ahead that\nyou do also just yeah go ahead sorry to\nbutt in a little bit it sounds like what\nyou had right there was a was a problem\nbasically just getting good training\ndata right absolutely I feel like that's\nthat's a big theme across a lot of these\npodcast episodes where uh 90% of it is\ngetting good training data and when you\ncan't get good training\ndata I mean you do the best you can but\nyou really just it's it's just that's\nall you can do\nyeah Absol you can't make miracles\nhappen yeah and also like labeling your\ncalls uh like for example we realize a\nlot of our calls are not even actual\nlike calls from the customer end they're\nsort of like workers at John Deere who\nare kind of working on the field okay um\nthey're not the real customers and of\ncourse they speak in a different tone um\nor way and it might seem negative or you\nknow it could be anything and the\nreality is actually you should not even\nbe looking at that call in the first\nplace and uh that tone is all relative\nto the type of people that are talking\nto one another yeah that's interesting\nthat's I mean I think there's a lot to\nbe learned from that right is is how do\nyou how do you when you're getting data\nwhen you're focusing on data how do you\nget the data you need and sometimes\nthat's really hard because it's usually\nall just put in one big bin in one big\nbox and they go here's our data and and\nyou're like\noh yeah got to sort through all this\nstuff and sometimes that's easy to do\nand sometimes that's like next to\nImpossible absolutely and all like the\nthe efforts to label the data that can\nbe yeah they might have tons and tons of\ncalls but we need people to\npainstakingly sit through them process\nthem and label them as red yellow or\ngreen um and that's not always the\neasiest thing to do we couldn't get a\nton of training data so that was also\nanother problem um but yeah that was my\nfirst internship okay uh I guess I won't\ngo over everything John Deere because\nit'll take too long but my second\ninternship um was is actually a um very\nuh natural language processing heavy one\nactually the last one was also natural\nlanguage processing because I had to\nprocess calls using natural language\ntechniques and I don't know if you've\nheard of TF IDF term frequency inverse\ndocument frequency I don't but I'll ask\nessentially trying to look at your calls\nto gauge um how important is a\nparticular word with reference to a\nparticular so like for example the word\nthe uh-huh is likely to appear in\neverything yeah so you look at also not\njust how frequently does the word the\nappear in a specific call and what\nsentiments associate with but how how\nfrequently does occur in all the calls\nand if there's like a particular word\nthat has that has you know is associated\nwith a bad sentiment and it's not like a\nnormal word we're like oh okay so that\nword is associated with bad sentiment\nthat's kind of how you know um things\nlike naive phase algorithms Works um\nyeah yeah yeah sorry I'll move on to\nyeah my next I love it I I know you know\nthis is great I one of the highlights of\nthis whole podcast is is a learning\nopportunity for me and so so I love it\nyeah so thank you very much for teaching\nme about a new term and a new\ntechnique yeah um sorry yeah so my next\ninternship was more natural language\nheavy and it this is the give context\nthis was after chat GPT came out the\nwhole generative AI boom and what they\nwanted to do I was working in Factory\nAutomation and they wanted to find a way\nto improve the completeness and accuracy\nof um machine defect documentation at a\nfactory um and to give context you know\noperators they be working on the\nassembly line and when they run into a\ndefect besides just figuring out what\nbase machine and what model and what not\nyou have to figure out look through like\nhundreds of thousands of parts and\nfigure out which part do you have the\nissue with and what exactly is wrong um\nand memorizing it well and that can be a\nvery tedious process so you wanted to\ncreate a sort of like a sub automated um\nsolution to this um this is also another\nimportant point that you don't want to\nalways completely rely on AI sometimes\nmerging it with human intelligence can\ngive you the best output over trying to\nf- tune and get perfect accuracy in your\nmodel um and what this approach\nessentially did was we used our bill of\nmaterials um and we put it into a vector\ndatabase type type of a setup and we\nstarted out broad it's a it was a\nhierarchical bomb so you have the broad\nlevel machine parts and the more\nspecific machine parts within those\nbroad levels and you kind of go through\neach level until you get to a specific\npart okay um and it's like a series of\nquestions like you'll be presented with\nthe top five parts to the like on the\nuser end and he or she would pick one\nand then you would get deeper um is it\nalmost like a decision tree to a certain\ndegree like or like a CL you know like\nlike where you know there's all these\nsort of if this part then this or or am\nI completely off on that with the\ndecision tree it's more hardcore ml\nwhere um well I guess in term if the the\none characteristic that's similar is\nnarrowing down your options uhhuh um\nhowever it's this one is very like um\nembeddings based so are the um the parts\nthey could converted to something called\nembeddings which is like a numerical\nversion okay of the words but also as\ncontext right it's not like if something\ncontains apple and this also contains\nApple then they're similar apple is also\nassociated with red so it has knowledge\nabout the English language and you do\nsimilarity searches with the users\ninitial description and you find the\nclosest thing but the nice thing is you\ndon't have to do it um you don't have to\nreally do it with all 200,000 Parts is\none they're categorized so you find the\nclosest bigger category yeah so how many\nof these parts um match it the most and\nwhat do most those parts fall into and\nthat's how you select the category oh I\nlove it I I I I love it because it's\nit's sort of like this divide and\nconquer strategy for identifying what\nthe real problem is uh rather rather\nthan trying to tackle everything all at\nonce you you've broken it down into\nsmaller pieces and I it makes a lot of\nsense to me I think it sounds fantastic\nthank you yeah um yeah yeah we that I\nguess it's really this is my favorite\nproject cuz um it's in a field that I\nreally like and it kind of brought in\nthe core of my major which is\nunderstanding how to integrate human\nintelligence with AI intelligence how to\nincorporate the human Centric aspect I\nlove that that my solution really really\nyou know was true to that um that you\nknow it's not just AI picking one out of\n200,000 Parts the user itself is picking\nit gradually and getting to the right\ndecision yeah and you're just helping\nthem make that decision fast fter and\nmore accurately absolutely I love it I\nlove it um uh so other questions I I've\ngot lots more questions um I'm curious\nso you say you're gonna uh start a\nmaster's program tell me about the\nmaster's program and also uh I'm also\ncurious as to um well we'll just start\nwith this tell me about the master's\nprogram master's program well the one\nthat I've applied to and hopefully get\ninto crossing my fingers um it's a data\nscience ones as I said and there's like\nyou know their classic beian statistics\ntype of courses um your ml type of\ncourses there's also like a cloud\ncomponent to it um and I can't remember\neverything on the top of my head but a\nvery like you get like the big picture\nof a lot of different you know from the\nstat side from the a ml side from the\ncloud side from the industry side how to\napply data science very is how I'd like\nto summarize it um and why I'm doing it\nuh initially my plan would be oh work\nfull-time only and figure out what you\nwant to do but I realized I didn't\nreally want to do classic software\nengineering I wanted to go into applied\nsciences um and their requirement was\nget a master's at most places so I was\nlike why wait if I know what I want to\ndo yeah um I might as well you know do\nthe job that I want and not wait if I\nalready know but and so the are you\ngoing to be you said it's all online are\nyou going to be doing this full-time\nthis master's program or are you going\nto be working as well or how's this\ngoing to work out sure yeah it's\nself-paced okay um so you can take one\ncourse or you can take three it's up to\nyou um and you can finish it uh whenever\nas long as you meet the deadlines so\nyeah I'm going to be working full-time\nthat is the plan um fantastic um and\napproximately how long do you think\nit'll take to complete the the Master's\nDegree um so usually it takes around 2\nyears however there's a way to transfer\ncourses in from undergrad if you took\nthem uh so I'm transferring in two\ncourses and I plan on taking it's an\neight course program so I should plan on\nbeing done in about a year so fantastic\ncongratulations that's exciting than you\num so I guess the question I have now\nnow that we know what we know where\nyou've been we know what you're what\nyou're doing and we know you're headed\nuh do you have any words of advice or\nthings that you learned along the way\nthat you think uh would be worth sharing\nwith other people yeah absolutely um\nwell I guess an overall advice that I'd\nlike to give college students in general\nis don't be\nlike be willing to experiment and try\nnew things um I think the best part\nabout my education that that it wasn't\njust hardcore CS it wasn't just is I\ntook like I had a I forgot to mention I\nhave a minor in CS so I got that whole\nyou know nice package I tried courses\nthat you know I might never apply ever\nagain like computer architecture but\nexposed me the things like lower level\nprogramming and how does programming\neven work um and how do operating\nsystems work and that kind of context\ncan really help you do better in the\ntech World um and also the importance of\nyou know having personal projects or\nbeing part of like for example I'm part\nof the disruption lab um tell me about\nthat yeah it's a tech Consulting\nacademic unit okay well yeah they' like\nto call an academic unit within the\ngeese College of Business um and you're\nessentially you're put into groups as\nsoftware engineers and you work for a\nclient usually like a startup or a small\ncompany um that's the usual thing and\nthat really helped me you know get you\nknow hands-on experience with like\nactual projects applying things in the\nreal world versus applying things in the\nclass are very very different yes yes\nthat's very true for sure yeah so that\nand actually how I even got into\ngenerative AI how I got my project at\nJohn Deere was because of disruption lab\num I was initially going to do like a\nvery classic analytics type of\ninternship for my second one but I'd\ncreated a um or with my team i' created\na live crypto search engine um that\nessentially yeah it was really cool like\nyou can like look up stuff so chat gbt\nyou know it has stuff 2021 and before um\nso you can't really get live summarized\ndata that's right so what we want to do\nis control the information Source uhuh\num to answer the questions that we're\nlooking up okay um in a search bar and\nthis is this concept is called retrieval\naugmented generation okay um and we use\nthis technology called Lang chain to do\nit um and we put apis on the back end\nthings like I don't know if you've heard\nof coin gecko or defi Lama and they fet\nthat information for you okay um so they\nhave all these like API calls it tries\nto figure out which API call to pick to\nanswer your question it takes the Json\noutput and summarizes into something\nthat a human can understand W that's\nfascinating um so there was one word\nthat that really threw me you said this\nwas\num some it was you said crypto sorry I'm\nso sorry so it was a crypto Search bot\nit's like basically you can ask any\nquestions that you want about crypto\nwhether it's like oh what are the top\ncoins in India or the US tell me it's\nmarket trends stuff like that that I\nmean that sounds fantastic I I love it\nso you've got you've got this large\nlanguage model or like a chat GPT type\nthing that rather than being limited to\ninformation prior to 2021 what you've\ntrained it to do is select one of these\napis from which it can get it can get\nlive information about Bitcoin or\nwhatever\ncryptocurrency someone is interested in\nand and you've trained it to select\nthose apis and then get the information\nand then when it gets the information it\nsummarized it which I this sounds\namazing I mean that's one of the coolest\nthings I've heard thank you yeah and the\nthing was at that time we didn't have\nthings like being AI that could give\nlive information so back then it was\neven cooler um but yeah it was but the\nfor me the interesting part was learning\nhow to use Lang chain and Lang chain\nbecame the foundation for my next\ninternship so um and that whole concept\nof being able to send text to chat GP\nbasically chat GPT um in an automated\nmanner yeah so so another thing I love\nabout what you're saying and this is\nactually kind of Echoes uh what we what\nI've we've heard in other podcast\nepisodes which is\nyou learn and then you apply so you\nlearned a new technique this Lang chain\ntechnique uh as part of this sort of\norganization on campus and then you took\nthat directly to your John Deere\ninternship and you applied it yeah I\nlove\nthat thank you I'm actually surprised\nyou know it's crazy I didn't I never\npicked the path of generative AI or\nnatural language processing one project\nsort of built on the other my first\nproject at John Deere I did natural\nlanguage processing um and daab liked\nthat so they put me on a generative AI\nproject and then John Deere liked it\nagain and put me on\na generative AI based project success\nbuild on success I love it I guess so\nabsolutely I'm fortunate uh that it\npaved out so nicely but I would like to\nsay on that note that if you don't know\nyour path it's completely okay it's okay\nto go to work for a bit figure out what\nyou want do only reason I'm doing my\nmasters once again is because I had that\nClarity and I know I'm not going to be\nable to get the job that I want if I\ndon't do this Masters so oh that totally\nmakes sense I love it well Kushi I want\nto thank you very much for uh taking the\ntime to be with us today and giving us\nsome advice and sort of just telling us\nabout your journey and your quest in\ndata science I absolutely love it uh so\nthank you very much for joining us\nabsolutely thank you so much for this\nopportunity Josh I once again I'm a\nmajor major sta Quest fan so this is\nlike surreal doing a podcast with you\nright now um but yeah thank you so much\nI really really enjoyed this\nconversation and would love to keep\nconnected with you he we'll do we'll\ndefinitely keep in touch",
  "transcript_chars": 25457,
  "ingested_at": "2026-05-15T10:54:36.889096+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 9982,
    "like_count": 161,
    "channel_id": "UCtYLUTtgS3k1Fg4y5tAhLbw",
    "categories": [
      "Education"
    ],
    "tags": [
      "Josh Starmer",
      "StatQuest",
      "Machine Learning",
      "Statistics",
      "Data Science"
    ]
  }
}