Vivan Poddar

Current freshman @ Berkeley. I spend most of my time teaching machines to design analog circuits. Turns out the hard part isn't the circuits, it's getting a model to reason about trade-offs instead of chasing a single number.

Right now I'm a researcher at the University of Utah's MicroElectronics Design Automation Lab (MEDAL), where I built ORACLE: a multi-objective reinforcement learning optimizer for analog and mixed-signal circuits with LLM-guided exploration. It swaps the usual scalar reward for a vector one, so the policy sees every spec at once and learns the Pareto front instead of a single point on it, and it generalizes to design targets it has never seen. Forthcoming at IEEE/ACM ICCAD 2026; it was also my ISEF finalist project.

Before this: two years at Doclens.ai, first building a PII anonymizer so legal documents could be safely handed to third-party AI tools, then coming back to overhaul the ClaimLens UI. Co-inventor on two filed patents from that work, including US-20250181617-A1. I also built Paradigms, a PDF RAG app for step-by-step academic support on the Gemini Live API, now used by 800+ students across several high schools, and I run Silicon League, a hackathon tournament with 3,000+ members across 20 organizations and $12,000+ in funding. Somewhere in there I put together a smartwatch-based screening platform for sudden cardiac arrest risk in youth athletes, validated it on 30 people, and put it on arXiv.

Along the way: ISEF Finalist, USA Computing Olympiad Gold, two-time National Medalist in Science Olympiad (and 80-something state and invitational medals I've stopped counting), and two first places in CS/Math at the Pittsburgh Regional Science and Engineering Fair.

Outside of that I'm into anything engineering, circuit design, and startups, and I like building things for STEM education that have an impact you can actually measure.

Talk to me about any of it, anytime.