AI summary by TwinMind
Human Stories in AI: Amy Finnegan
StatQuest with Josh Starmer
28m 25s
Educational Background and Early Career Development
- PhD in 2016 from Duke University, public policy program
- Concentration shifted from political science to sociology/demography
- Data science wasn't established field yet
- Advisors recommended using secondary data vs. collecting own data to graduate
- Learned skills now common in data science: secondary data merging, quasi-experiments, A/B testing
- Multi-disciplinary program exposure
- Committee required economist, sociologist, political scientist
- Realized common thread was regression, data, descriptive statistics
- Central limit theorem consistent across disciplines
- Focused on learning methods well to address any problem
- Self-described as "mile wide and an inch deep" - interested in many things
- Found loophole by focusing on methods rather than specific problems
Postdoc Experience at Duke Global Health Institute
- Transitioned from secondary data to field data collection
- Postdoc advisor automated scripts for data collection/cleaning in R
- Key advice from computational political science class (taught by Microsoft developer-turned-PhD student):
- Social scientists who can program are more impressive than developers who program
- Programming skills help solve problems from any direction
- Postdoc designed to groom for NIH-supported research scientist track
Research Focus on Demographic and Health Surveys
- Discovered demographic and health survey data for dissertation research
- Used contraceptive calendar and maternal mortality modules
- Described as "gnarly" datasets people avoid due to difficulty
- Saw complexity as "Big Data opportunity"
- Led Big Data for Reproductive Health group at Duke for 4 years
- Worked with Duke students using Big Data methods on contraceptive calendar data
- Dissertation focused on maternal mortality module from demographic and health surveys
Transition to IntraHealth International
- Research scholar position at Duke Global Health Institute (2019)
- No clear career pathway from research scholar to professor
- Had 5-year plan and career development plan
- Connection through colleague David Pogeni (informatics specialist, ~70 years old)
- Had "budget dust" - leftover fiscal year money for innovative projects
- Saw opportunity to connect siloed health data systems using machine learning
- 2-month consulting project became extended job interview
- Initially hesitant about job posting, felt overqualified requirements
- Accidentally negotiated good deal when director asked "what do I need to do to give you this job"
- Hired as senior data scientist in September 2019
Key Career Advice for Job Applications
- Focus on "here's what I can deliver for you" vs. "this would be a great opportunity for me"
- Build trust through small projects leading to bigger opportunities
- Show mutual benefit and time investment required
- Demonstrate results from previous projects
Current Role and Major Projects at IntraHealth
Ready (Rapid Efficient and Data Driven Implementation) Project
- Unified data from all projects feeding to technical experts, executive team, board
- Built system over several years
- Led technical working group including:
- Technical advisors (define what goes in system)
- M&E people (define indicators, measurement)
- Data science team (build structures, data pipelines)
- Created PowerBI dashboards for data visualization
Tanzania Voluntary Medical Male Circumcision Project
- Challenge: Circumcise more men with fewer dollars, targets increasing while budgets flat/decreasing
- Problem: Fewer eligible men over time (procedure only needed once)
- Mobile testing sites and roving health provider teams often went to wrong locations
- Solution using multiple data sources:
- Facebook's 1x1 kilometer high-resolution population density maps (satellite data)
- Demographic and health surveys for circumcision rates by age
- PEPFAR data on circumcisions performed quarterly over several years
- HIV prevalence data from testing
- Created aging algorithm: subtract already circumcised, add new 14-year-olds annually
- Result: Interactive map showing optimal locations with high HIV prevalence and uncircumcised men
- Project was Tohara Plus funded by CDC in Northern Tanzania
Indonesia Desa Siaga Program Evaluation
- "Alert Villages" program creating disaster response network
- Designated village first responders for disasters, maternal complications
- "Suami Siaga" (alert husbands) - volunteer ambulance service
- Evaluated program effectiveness using facility and delivery data
Technical Tools and Approach
- Primary tools: R (free/open source), QGIS (geospatial work), PowerBI (dashboards)
- Microsoft shop integration with PowerBI
- Geospatial data work requires specific R libraries
- Data linking challenges: matching facility/location names across datasets
- IntraHealth's "Gopher" tool (Global Open Facility Registry) for data matching
Current AI/LLM Project
- Using large language models (ChatGPT/GitHub Copilot) for open-source health software development
- Problem: 10-15 year old open-source health information systems lack funding for maintenance
- Tools include: facility registry, identity management, supply chain, health worker management, electronic health records
- Originally funded under non-permanent projects
- "Everyone assumes someone is watching" but no one takes responsibility
- Solution: LLM assistance for software development on IRIS (health workforce planning software)
- GitHub Copilot ($10-20/month) provides context-aware code assistance
- Goals: accelerate adoption, make software more robust with less funding
- Intern working on project using LLMs for documentation, bug fixes, test writing
Final Career Advice
- Cannot work alone - need full team approach
- Data scientists have Venn diagram of technical/hacking skills and statistics, but need complete team
- Get buy-in and create value for others to benefit from solutions
- For students: create project portfolios, use GitHub, document code well
- Show potential employers you can "fit in and start producing value on day one"
