My agents do real work, like calling a hospital's billing office or drafting a client change order, but the math is done in code and nothing goes out until a person approves.
Penn State CS '26 · 5× hackathon winner · available January 2027 for Forward Deployed Engineer and AI Engineer roles · open to relocate
Building now: PrismAI, an approval-first meeting agent. 14 beta users, 100+ meetings analyzed; agency release in December.
| Project | What it does | Recognition |
|---|---|---|
| PrismKeep | Six agents run a founder's follow-ups; none tells an investor something a customer contradicted | 🥇 1st overall, SF Startup Challenge 2026 |
| FireSight | A cited 3D arrival brief for fire officers, from 88 San Francisco public data sources | 🥇 Track 1: Public Data APIs, super{set} SF |
| Fish Market | AI agents compete for your order, every listing is screened for scams, and big buys wait for your OK on a phone call | 🥇 Best Use of Fetch.ai, Cal Hacks 13.0 |
| BetterBill | Finds errors in hospital bills; a voice agent calls billing once you approve | 🥇 Visa track · 🥇 Photon: Best Voice Agent, LA Hacks 2026 |
| Terrametric | Field-level drought risk and irrigation plans for farmers | 🥇🥇 + 🥉 at BitCamp 2026 |
| arXiv:2607.21641 | Tool-guided repair for LLM-written C: 83.7% fewer CodeQL errors | First author · 2nd Place, Penn State Exhibition |
| Match Day | Runs a racket tournament from one screen, live scores on every phone | Live |
| Vytal | Coaching platform where every score is deterministic and explained | Controlled beta |
| MVR | Verified software modernization: every claim about a legacy codebase links to evidence | Proof of concept · 2,320 tests |
| Yinzone | Where new housing can go in Pittsburgh, across 142k parcels; I built the cited AI chat | AI for Housing Hackathon 2026 |
Also built: FilingWatch (SEC 8-K monitor) · CinemaChat (agentic movie recommendations) · TransitGuard (airport bag-transfer risk) · PaperTrail (learn a field from its foundational papers)
Before this: SWE intern at Sherlock AI and Comcast, where a 40-person team adopted my agent pilot and cut ~40% of its planning work.




