Advay
Bhattacharya

So What

Door 5001. That number is in my college application because it was the one where the work started to feel real. My supervisor and I had been walking neighborhoods for a congressional campaign, and somewhere past a thousand doors I stopped thinking of the job as voter outreach and started thinking of it as a data problem with a face on it. A Vietnam war veteran on a quiet street who asked us to stay and talk about the future of the country. A family at a Diwali event at the local football stadium who had just arrived in America and signed up to vote for the first time, then got their sons involved. A community in a trailer park that had been marked as nonvoters not because they were disengaged but because the targeting data had never looked closely enough to find them.

At UHG, the same gap shows up differently. Over 3,200 accounts worth $160M had been routed to the wrong vendors, invisible in the system until a SQL pipeline caught them. The for-profit health plan client from my second internship sends its revenue directly to the Medicare communities it serves, often elderly patients who depend on that income stream being accurate. In surveys of providers, risk adjustment operations are cited above premiums as the primary revenue-generating arm. The compliance reporting I automated wasn't an internal operations problem. It was the operational reliability that keeps that income flowing. The work looks like Power BI and Power Apps. The reason it matters looks like those people.

The less visible change is what this work did to how I think. The trailer-park community gave me an assumption I've carried into every project since: any dataset has a population it's systematically missing, and the real work is usually finding them. At UHG, I wasn't asked to look for the 3,200 misrouted accounts. I went looking because that assumption had become automatic. Mock Trial added a different kind of discipline. Learning to hold cross-examination without improvising past the case materials, staying inside what the evidence actually supports rather than what feels plausible, turned out to be the same constraint that matters in regression work. I didn't connect those two rooms until I was deep into a multicollinearity problem and realized I was doing the same thing I'd been trained to do on a witness stand. These experiences didn't just teach me skills. They changed the lens I reach for first.

Now What

I came across Han van der Maas's mutualism model in Google Scholar in high school while procrastinating on a debate case. The idea stayed with me: math, language, and spatial reasoning develop strong correlations over time not because they share some common factor, but because growth in one actively enables growth in the others. Four years competing in Mock Trial built the same analytic discipline I use diagnosing multicollinearity in a regression model. Decomposing a witness's credibility under cross-examination and isolating a collinear predictor are, structurally, the same problem. My Honors contract in Math Stats II applying statistical inference to network data grew from that same instinct to map how things connect rather than study them in isolation. The degree-plan visualization on this site is a map of how courses and skills actually connected for me, with edges drawn where one opened the next.

The framework I keep returning to for understanding any new domain is Bloom's Taxonomy: remember, understand, apply, analyze, evaluate, create. I use it to locate exactly where I am in a subject rather than treating mastery as binary. MECE (Mutually Exclusive, Collectively Exhaustive) is how I break any domain into its components: no overlap, no gaps, each piece doing its own work. I use AI now not to generate answers but to probe which level of Bloom's I've actually reached on a topic, then push to the next. The combination has let me move across political science, statistics, healthcare operations, and network science without losing the thread between them. The path ahead points toward healthcare data science, not as the obvious next step from UHG, but because the problems there have the same structure as everything I've worked on: operational gaps with real people on the other side of them.