I build full-stack AI tools, data products, and practical software systems.
Lately, that's meant an award-winning agentic iOS safety app and a multi-agent urban planning system built on six public datasets.
iOS app that estimates impairment by measuring deviation from your own sober baseline — reaction time, balance, memory — using an agentic AI examiner, not a BAC guess. 3rd place, AI for Social Good: Hack with MLH & DigitalOcean (Jul 2026).
Tool-calling loop grounds every verdict in specific test results: retrieves baseline, requests targeted tests, analyzes deviation, and updates a confidence score.
Agentic code-risk analyst: point it at a real repository, click a function you’re about to change, and a graph walker computes exactly what’s downstream — then an LLM agent explains the risk, grounded in the real source. Built for JacHacks SF 2026.
Every claim in the UI is labeled by source: affected functions and suggested tests are computed from real call-graph traversal, while the risk narrative and severity are AI-generated — the agent reads real source via a tool call before it answers.
Multi-agent AI system that runs parallel scenario analyses — climate, housing, accessibility, urban design — for any real address.
Integrates + cleans 6 data sources: Census ACS, FEMA NFHL, Open-Meteo, 511 GTFS transit, NLCD tree canopy, and Google Maps. Supports multi-variable sensitivity analysis across planning dimensions.
| Category | Current | 2040 | 2075 |
|---|---|---|---|
| Climate | 62 | 72▲10 | 84 |
| Accessibility | 94 | 96▲2 | 97 |
| Housing | 78 | 84▲6 | 91 |
| Overall | 78 | 84▲6 | 91 |
Voice-first automation platform on Meta Ray-Ban glasses — one spoken command triggers multi-step workflows across 12+ services, including Gmail, Slack, Notion, and Fetch.ai Agentverse. Built as BarelyAtWork at LA Hacks 2026, in a team of four over 36 hours.
Constrained Gemini pipeline with a validate → repair loop converts voice transcripts into structured workflow JSON and eliminates invalid parameters.
Modular overhead gantry and cart system for semiconductor cleanroom material handling, co-designed to ISO 14644 standards. 2nd place, ASME UC Berkeley CADathon Spring 2025.
Trade-off analysis across layout alternatives; materials selected within a $2,300 BOM to meet cleanroom and conveyance requirements.
Kepler photometry pipeline that detects, characterizes, and stress-tests a real exoplanet transit signal — built with a 7-person team for Astron 9 (Introduction to Python for Astrophysics).
A Box Least Squares periodogram recovers a 3.52-day period at power ≈570 against a ≈10–50 noise floor, matching the published orbital period of Kepler-8b — an independently confirmed exoplanet, recovered from raw NASA brightness data alone.
Designed and ran measurement studies across 15+ experimental configurations, isolating performance factors and failure modes. A/B-tested ML policy variants to quantify trade-offs, and processed large-scale operational datasets in Python/Pandas for engineering stakeholders.
Mentored 15+ high school seniors one-on-one across two application cycles; facilitated operations at the 2024 USC and 2025 Princeton Young Leaders Summits. Mentees admitted to Yale, Columbia, Brown, Tufts (QuestBridge), USC, WashU, and UC Berkeley, including a Gates Scholarship recipient.
Managed training camp operations for 4,000+ attendees, coordinating logistics against team and fan-engagement goals. Streamlined internal workflows by communicating across executive, coaching, and event stakeholders.
Managed pricing, inventory, and fulfillment for an e-commerce resale business generating $100K+ in revenue. Scaled to international markets while sustaining 100% positive feedback and eBay Top-Rated Seller status.
Fresh off building BuzzBuddy at an MLH × DigitalOcean hackathon — now polishing this website you're looking at.
A summer course learning about Python for astrophysics.
Internships in software or hardware, learning more about culture, and developing new skills.