{
  "video_id": "m8fvdRZb5CE",
  "channel_slug": "statquest",
  "channel_handle": "statquest",
  "title": "Human Stories in AI: Amy Finnegan",
  "duration_seconds": 1705.0,
  "url": "https://www.youtube.com/watch?v=m8fvdRZb5CE",
  "upload_date": "",
  "transcript": "hello I'm Josh starmer and welcome to\nhuman stories in AI with stack Quest and\nlightning AI in this series we'll hear\nabout the career journeys of passionate\nAI experts from their humble beginnings\nto conquered challenges will be inspired\nby the realworld experiences of\nprofessionals thriving in the ever\nevolving AI\nlandscape human stories and AI is\nbrought to you by lightning AI code\ntogether prot type train and deploy AI\nweb apps all from your browser with zero\nsetup personally I love lightning AI\nbecause 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 Dr\nAmy finnean the deputy director of data\nscience at intra Health International\nAmy is a demographer and data scientist\nwith over 10 years of experience working\nin global Health development and data\nscience in emerging economies on four\ncontinents Amy is also an adjunct\nfaculty member at Duke University's\nGlobal Health Institute so without\nfurther Ado Amy can you tell us about\nyour journey to where you are right now\nat intra Health how did this all\nstart yeah absolutely um I uh you know\nwon't go back to all the way back but um\nI got my PhD in 2016 from Duke\nUniversity and I was in the public\npolicy program with a concentration in\nuh started out in political science\nended up in sociology demography and at\nthe time I was doing my PhD data science\nwasn't really a thing um but in the\nprogram um all my advisers told me you\nknow don't collect your own data because\nyou'll never graduate so we ended up\nhaving to\nuh use secondary data merge data sets\ntogether ask interesting questions do\nquasa experiments run AB tests like all\nof that stuff that is now commonly part\nof what a data scientist does is what I\nwas learning in grad school um before\ndata science was even a thing um and\nthen within that as well we were in a\nmulti-disciplinary program so you know\neveryone your committee has to be um an\neconomist a sociologist and a political\nscience\nand they are all experts in their field\nbut they all approach the same question\ndifferent ways so I realized pretty\nquickly that the one thing we had in\ncommon was regression and data right and\ndescriptive statistics and like the the\ncentral limit theorem doesn't change\nbased on which uh discipline you're in\nso I really doubled down on learning\nmethods really well and um coming from\nthat interdisciplinary or\nmulti-disciplinary program I felt like I\ncould kind of address any problem using\nthose same methods and I also you know\nwasn't interested in like one problem\nspecifically so I was probably you know\nI don't know if you've heard this saying\nbefore but like a mile wide and an inch\ndeep okay sure that's not allowed really\nor not encouraged in PhD program right\nthey want you to be the the expert in\none specific thing and I was interested\nin so many things that I found the\nloophole was like being interested in\nmethods okay awesome so yeah and then I\nwent to uh the Duke Global Health\nInstitute and I did a postto there in\npublic health where I actually was\nstarting to work on projects where we\nwere collecting data in the field um I\nhad a really great postto adviser who\nwas automating all of his scripts to\ncollect and clean data in R and I kind\nof sought him out because I really\nthought that those were skills that I\nreally wanted to have um and I got this\npiece of advice kind of Midway through\nmy PhD program in a class called\ncomputational political science which is\nprobably now just called data\nscience and they said it was taught by\nsomeone who had worked at Microsoft and\nthen came as a developer and then was in\nthe PHD program in political science at\nDuke and he said if you're a\ndeveloper and you can program everyone's\nlike great you're expected to program um\nbut if you're a social scientist who can\nprogram then everyone is impressed and\nthey're like this is like an additional\nskill that helps you stand out and then\nyou can solve problems um that are\nyou're coming from any direction so\nthat's why I really um focused on that\ndid you have a follow-up question well I\nwas just going to ask uh and I think you\nmay have answered this already but so\nyou were in this PhD and then you moved\non to a postto how did you choose that\npostto was it specifically to learn how\nto program or or or how what how did you\nmake that decision yeah it was to go\nfrom using secondary data to Public\nHealth where you're collecting your own\ndata and to really be groomed on that\ntrack of like being an NIH supported\nresearch scientist and designing\ninvestigator initiated studies um and so\nI think the biggest compliment I've ever\ngotten is that I'm creative so I really\ndoubled down on that right like a lot of\npeople have skills you can be super good\nat something and follow all the rules\nbut to kind of push the envelope you\nreally do need to be\ncreative other people CH grin at times\nI'm\nsure but how did that work in as a\ngraduate student that that did that\nchafe you because they didn't they\ndidn't want you to gather your own data\nor um I really loved the demographic and\nhealth survey like I can still remember\nwhen my advisor said some of these\nquestions that you're asking you could\nprobably answer with the demographic and\nhealth survey and I was like okay look\nlet me check it out and that's what I\nused to write all the papers in my\ndissertation That's the basis of like\ncreating this big data for Reproductive\nHealth um bass connections group at Duke\nUniversity for about four years we were\nworking with Duke students to use Big\nData methods on the contraceptive\ncalendar from the demographic and health\nsurveys and then my dissertation was on\nthe mat using the maternal mortality\nmodule from the demographic and health\nsurveys so these two kind of gnarly we\ncan say data sets that people don't use\nbecause they are so hard to Wrangle we\nsaw that as like here's the Big Data\nopportunity and we can um use these new\nmethods that are becoming um more\npopular to answer questions that we have\nabout reproductive Health oh I love I\nlove this so let me just make sure I get\nthis straight though uh it sounds like\nso this data set is Big complicated hard\nto\nuse and you saw that as an opportunity\nyeah absolutely I love that I absolutely\nlove that fantastic Okay so we've got\nyou as a postto learning\nR where do things go from\nthere yeah so I finished up my postdoc\nand then I had a job as a research\nscholar at the Duke Global Health\nInstitute and everyone will tell you\nthere is no track there's no like track\nfrom research scholar to like research\nProfessor or you know there's no real\ncareer pathway in that job but I had\ngone from a postto at Duke and really\nwanted to stay in the triangle area so\nI'm based in Durham North Carolina and\nuh kind of this was the The Next Step\nfrom post do it kept me at Duke\nUniversity um where I would had already\nkind of like figured things out I felt\nlike seven years you know I kind of\nunderstood what was expected of me as a\nresearcher and like I know where I'm\ngoing I had my my career development\nplan and my Five-Year Plan and all of\nthat um and then someone I worked with\nactually on big data for Reproductive\nHealth had gone to work at intro Health\nInternational where I work now and they\nhad uh at the end of the fiscal year\nthey would always have some what they\ncall Budget dust like okay left leftover\nmoney that they could use before the end\nof the fiscal year right like the sound\nof that yeah right and it's it's dust\nit's not like big project material but\nthey would use it for kind of innovative\nprojects that they could see if they had\nany legs right and so this guy David\npoeni if you're listening\nDavid he said that um uh he had they had\na bunch of data uh he was an informatics\nguy um I think he was pushing 70 when um\nI ran into him and he had seen like how\nhealth informatics had unfolded in the\nUS and then also supporting Global\nprojects and he was like we've got all\nof this data it's not connected like\nit's siloed into different Data Systems\nbut the methods exist to link all this\ndata together and then the machine\nlearning methods exist to like make\nsense of it right so that was his\nproject proposal and through this person\nI had worked with at Duke they found me\nand had me as a consultant for um two\nmonths uh and it turned out to be a\ntwo-month long job interview so I didn't\nknow that uh they were GNA hire someone\num at the end of this so when my when\nthe job ended actually this is a funny\nstory um so the job ended they posted\nthis job they sent it to me and I was\nlike you know I don't think I'm who\nyou're looking for like I think this is\nlike requesting too much experience and\nlike too too much of all of that and\nthey were like no just like come in for\nthe interview and I was like okay sounds\ngood so I came in for the interview and\nI met that director and David was there\nand other people on the team and the\ninterview went really well they followed\nup for the second interview and I was\nlike you know like I feel like I I\nfinally figured out what I'm doing at\nDuke and um this would be a big change\nfor me and uh I had just got off the\nwaiting list for parking it had been two\nyears that I was on this waiting list I\nwas like oh decisions um so the director\nof digital Health at that time called up\nand he was like what do I need to do to\ngive you this job so I accidentally\nnegotiated a pretty good deal for myself\noh that's awesome I love this\ntechnique yeah it was definitely an\naccident but I I kind of bumbled my way\ninto this ex really great job I love it\nI love this\nstory um well I mean to be honest it\nsounds like uh it sounds like you were\ngiven an opportunity to do this sort of\nConsulting and you you took it you\ndidn't that was the key right if you\nbumbled on that and been like I don't\nknow if I want to do this they you may\nhave never heard from them again but\nbecause you you took that opportunity\nand you went with\nit they fell in love with you and they\nwere like we have to get you and that's\nI think there's something to be said\nabout that you know just sort of going\nout on a limb and and just saying oh I\nthink this will be fun and let's see\nwhat happens yeah and I you know I gave\nup my weekend for a couple months to\nwork on this\nduke it and Duke also rules how much\nConsul you can do and so it to fit\nwithin the rules of that but yeah um\nmade it work out really fun yeah we made\nit work and I love it yeah and then they\nhired me as a senior data scientist and\nthen uh that was September 2019 yeah\nthat brings me to something that I did\nwant to say like advice for students\nfrom reviewing a lot of student\napplications for our bass connections\nteam reviewing a lot of like like I've\nhired three data scientists now at inal\nlike reviewing cover letters and all of\nthat the one the people who say like\nthis would be a great opportunity for me\nare less appealing than those people who\nare like here's what I can deliver for\nyou okay okay yeah nice so that's very\nimportant distinction as people are\napplying for jobs oh I like that yeah\nit's uh it's it's it's basically doing\nthe exact same thing you did in Africa\nlike here's what we can do for you bam\nyep selling it right and here's how we\nmutually benefit and this is the time\nit's going to take and getting a little\nbit of trust right and building up your\nprevious projects you can show them the\nresults you've had and then they're\nwilling to um trust you for the the\nbigger dollars right and that's how the\nUganda Consulting project happened is\nthat that turned into bigger dollars\nfrom that same project that ended up\nfunding a lot of my time in that first\nyear because we were given the runway to\njust like see what can happen and then\nwhen we were able to solve a problem the\ncheve a party of that project was like\ncan you solve this problem too uhhh this\nis the one that I really need solved so\nso you you said you said a word that I\nthink is super important to just data\nscience in general uh which is trust\nbuilding up trust um I can think a few\nthings more critical to sort of the role\nof a data scientist than establishing\nand building trust um because you know\nwhat we do can somewhat seem like magic\nyou know like machine learning AI all\nthese things there's a lot of buzzwords\nin our field that I think within the\nfield we know what we're talking about\nwe know what these things are but\noutside of the field I feel like a lot\nof people are\nintimidated uh or or they're or or you\nknow or just like they don't know what's\ngoing on and there's that fear of the\nunknown or the or the fear there's just\nfear I think um yeah it's like every\ndollar we spend the opport Unity cost of\nspending on something that might not\nwork is it's a risk that a lot of people\naren't willing to take yeah and I think\nwhat you said was how you start you know\nsimple and you build up trust over\ntime uh both by like completing tasks\nand showing utility but I think also a\nlot of that is uh has to do with clear\nCommunications in terms of like how this\ncan benefit you yep and also walking\npeople through what you're doing and\nmaking it intuitive for them because a\nlot of machine learning it's it's an\nartificial intelligence it's supposed to\napproximate like how the brain works\nright like in like creating that\nintelligence so if you can get those\nreally simple examples that help people\ngrasp that intuition then um yeah they\ncan kind of wrap their heads around it I\nreally do think that's one of the Gap\nareas in our field is that it happened\nso quickly that um you know software\ndevelopers started doing data science\nand in ouri field strategic information\nand m  people started doing data science\nbut the Gap I think was teaching the the\ntechnical people how to like get the\nintuition and be able to interpret these\nthings and then ask for it right because\nit's only as good as the questions\nyou're asking from it as well and kind\nof thinking about their specific um\nproblems and how data science could be\nused to solve them and I think\nupskilling some technical people would\nbe a really good investment yeah that\nmakes sense makes sense to me so could\nyou tell us a little bit about what\nyou're doing at interel right now yeah\num so I've been here for four years and\none of our major projects is called\nReady rapid efficient and data driven\nimplementation um because there was a\nneed to um kind of unify all of the data\ncoming in from our our projects and how\nthat was getting fed up to our um\ntechnical experts at the global level\nand then up to the executive team and up\nto the board so I spent the last couple\nyears building that system um helping uh\nrun a technical working group where it's\nnot just data science building it it\ncan't just be data science um you know\nwe have the technical advisers who tell\nus what should go in there we've got the\nm  people who tell us how to define\nthose indicators and make sure they're\nmeasured well and then the data science\nteam can build those structures and data\npipelines to bring them in and then\ndisplay them powerbi dashboards which is\nwhat we do so that's been our major\nproject and our probably highest\nvisibility project at the organization\num one of the really cool projects we\nworked on um in Tanzania was on um\nvoluntary medical male circumcision so\nwhat we're seeing is that donors are\nasking us to circumcise more men with\nfewer dollars right so your targets are\ngoing up your budget's either staying\nthe same or going down and you still\nneed to reach your targets okay so what\nwe learned after implementing for a\nwhile is that the the more men you\ncircumcised the fewer men there are to\ncircumcise right you only need to do it\nonce um and so we had older population\ndata from um the Bureau of Statistics in\nTanzania that was at a higher level so\nit was at like the the region and the\ndistrict level for example so we had a\nmobile testing site and we had kind of\nroving teams of Health Providers who\ncould go to health sites and and perform\nthe circumcisions and then upskill\nproviders while they were there um but\nwhat ended up happening is that these\nteams would go out and stand around\nbecause they didn't go to the right\nplace there weren't as many people as\nthey thought they were there were you\nknow the even these districts and um\nregions can be so big that even the the\nhealth officers there might not know\nwhere all the people are especially if\nthey're um populations that are moving\naround a lot like pastoral communities\nand things like that so this was a\nrequest from the Project Director of\nthat project it was toara plus funded by\nthe CDC in Northern Tanzania and she\nsaid look look like I'm having this\nproblem we we need to go more granular\nso we found some data actually from\nFacebook created these 1x1 kilometer um\nhigh resolution population density maps\nand put them out into the wild\nuh so we could see which it had been\nsatellite data that they had collected\nfor free and then estimated where people\nwere by different age groups um so we\nhad that which we could aggregate to the\nthe district level and then also below\nthe district to the W level and so we\ncould show where the people are where\nthe men are right and then we could use\nthe demographic and health surveys which\nyou know I love um to estimate you know\nif we've got this many 15 to 19 year\nolds the DHS is telling us that by age\n15 x% of them are already circumcised so\nlet's take the number we have and reduce\nthat by the percent we think are\ncircumcised and then pepar who funds\nthese programs it has been collecting\nlike the number of circumcisions done\nevery quarter over the last several\nyears so then we could also decrement\nfrom that number the circumcisions\nalready done and then add in the 14y\nolds right and like cuz people age one\nyear at a time that's the best thing\nabout\npopulations um so then we were able to\nOverlay that with data we had on HIV\nprevalence which we got from the testing\nand the positive test from pepar and the\nfirst time I made that map like popped\nup on my screen and it just lit up like\na Christmas tree right it's like here\nare the wards we need to go to that have\nhigher than average HIV prevalence and\nwe can sort those by the number of men\nthat we need to circumcise so we know\nexactly where to go W so yeah just to\njust for my benefit uh not being in the\nhealth uh business it sounds like uh so\nit sounds like the the end uh was or the\nthe purpose was to reduce HIV rates yes\nand and the way you could go about it\nwas uh increasing the uh proportion of\nthe male populations that's been\ncircumcised and so that's what you guys\nwere were were tackling uh which makes\nsense so you're trying to get it before\nyou know at an early you know trying to\nlike prevent it before it happens yeah\nit was a prevention project for sure\nyeah that sounds fantastic um can you\ntell us so this data sounds really cool\nuh is it I mean how was it working with\nthe data was it is a huge cumbersome\ndata set or is it how do you how do you\nwork with it yeah it wasn't too bad like\nyou need to know some of the basics of\ngeospatial data we were working with it\nin R so you have to know the library\nthat you can work with geospatial data\num and then there are some really good\npackages for summarizing those pixels\nessentially like you give it the the\nward boundary and then you sum up the\npixels in that boundary and that tells\nyou how many people there are and then\nyou merge on the data from the\ndemographic and health surveys or from\npepar that's also at the ward level so\nyou do have to know like what are those\nidentifiers that you're going to use to\nlink the data together\num and then the issue can be that one\ndata set says this is like a sunshine\ndistrict and the other data set says\nthis is sunshine a district and like\nwhich one is correct you know and is\nthis really the same place so in health\nplug for INT Health has a tool called\ngopher which is the uh Global open\nfacility registry that helps you link\nthose two lists together using some um\nmatching Theory so there there are ways\naround it but it was very similar to\nsomething I did for my dissertation\nwhich was using data from Indonesia to\nmap on um deliveries and facilities and\nto evaluate this program called um Desa\nYaga where uh they had this kind of\nRippling implementation of the program\nand we want to see if things were better\nafter the program what was the program\nsupposed to accomplish daaga means alert\nVillages so it was supposed to have a a\npost in the village that didn't exist\nbefore or was kind of like repurposed to\num like if a disaster struck like the\nbig tsunami people were designated to be\nthe First Responders um someone would\nvolunteer to be like you can call me and\nI will be the ambulance like if a woman\nhas complications so that was the suami\nsiaga that's the uh alert husbands\nprogram so it was really about um making\nsure that people were aware and planned\nfor things that could go wrong just like\ndisasters in general disasters maternal\ncomplications and childbirth um probably\nmalaria Deni fever co could activate\nthis network of Health posts to provide\ninformation\ncool um that sounds very helpful um uh\nso it it sounds like you do a lot of\nyour work in R do you do all of it in R\nyou said you got this for doing\ndashboards you guys use powerbi yep we\nuse powerbi for our dashboards because\nwe're a Microsoft shop so it nests\nreally well with all of our other tools\num we use qgis for our geospatial work\nbecause it is free okay that works good\nfor me R also because it is free and\nopen source so are the people that we\nserve uh you know they can't they can't\nafford support even the charity pricing\nthat we get from Microsoft so yeah if\nyou're looking at like what do I spend\nmy dollars on licensing is not at the\ntop of the list when you've got um women\ndying in child birth right yeah yeah\nthat makes sense um very cool\num are there any other projects going on\nright\nnow um we're trying to get one project\nstarted um because\nuh I'll go I'll go on my open source\nsoftware\ntangent so uh about 10 or 15 years ago\nwhen um open source software started to\nbecome developed for Health Information\nSystems uh to replicate like what we\nhave in in the US to make that open\nsource so that it would support a whole\ncountry's health information system like\nthe facility registry uh identity manag\nmanagement Supply Chain management\nHealth worker management electronic\nhealth work electronic health records um\nall of those tools are they're open-\nSource versions of those tools available\num but they were produced or initially\nfunded under projects that were never\nintended to be permanent so you've got\nthese pieces of software that um you\nknow everyone assumes because they're\nopen source like everyone is watching\nand there's a community but what it\nreally means is that no body is taking\nresponsibility and there's no funding\nfor um improving these core tools\nunfortunately um so one project we're\nworking on is using large language\nmodels like chat GP um to develop our\nsoftware right because it's it's open\nsource so we can just say like chat GPT\nread our software and then you know how\nyou know vue.js and no. JS and open\nsearch um help us document this code\nright or look at this bug and suggest\nhow we can fix it or um write tests for\nus right because our developers don't\nalways have room or time in the budget\nor things move so quickly or priorities\nchange that sometimes testing falls to\nthe Wayside but it's a really important\npart of software development um so we\nhave an uh an intern now working with us\non using large language models to\nimprove our our Iris Software which is\nHealth Workforce uh basically headcount\nplanning software for\ngovernments so yeah and do you run your\nown version of chat GPT or do you just\nuse the we're using github's co-pilot\nright now uh it is like 20 bucks 10 or\n20 bucks a month um but it's a context\naware so you can have this um\napplication where you open up all of the\nsupporting files for IRS and then it can\nread through any of them um and then\nit's also polling from um\nuh what it's learned from the web about\nsoftware development and like VI and\nnode and and all of that so we're really\nhoping that it can accelerate adoption\nof the software that it can make them\nmore robust with less money um sounds\ncool sounds very cool well uh to wrap us\nup I know you already gave us one bit of\nadvice uh that you've learned along the\nway but do you have anything else that\nyou've learned that you think would be a\nbit of wisdom that you could share with\nour\naudience yeah one piece of wisdom I\nthink yeah realize you can't do it on\nyour own you need a team of people to do\nit the data scientist has that kind of\nvend diagram of skills of like technical\nhacking skills and statistics but you\nstill need a full team of people to make\nthis work so yeah get get Buy in create\nvalue for other people so that they will\nbenefit from your solution um students\nshould be doing projects and creating\nportfolios so that they can show\npotential employers um what what they're\ncapable of doing you know put everything\non GitHub document your code really well\nshow show teams that even at an entry\nlevel like you can fit in and start\nproducing value on day one I love it\nthank you very much Amy it was great\nspeaking with you today and uh have a\ngreat rest of your day great thank you",
  "transcript_chars": 25453,
  "ingested_at": "2026-05-15T10:53:48.663729+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 5991,
    "like_count": 88,
    "channel_id": "UCtYLUTtgS3k1Fg4y5tAhLbw",
    "categories": [
      "Education"
    ],
    "tags": [
      "Josh Starmer",
      "StatQuest",
      "Machine Learning",
      "Statistics",
      "Data Science"
    ]
  }
}