Machine learning, the craft of building product, and the nerve to start things. The edges are just as fun.
Shipped a production banking demo in Knot's developer docs that gave card-issuer stakeholders their first hands-on look at card-switching and transaction-data workflows, and engineered an end-to-end API reference integration spanning auth, the Web SDK, webhooks, and SKU-level sync. Mapped prospect and developer behavior across Knot's docs and funnel by analyzing engagement data in BigQuery.
Advancing AI-powered student support through Student Information Technology for 10,000+ Carnegie Mellon students. Run the campus AI assistant, using student feedback loops to sharpen answer quality and resource accuracy, and refined AI resource discovery across 350+ student organizations.
Advancing crypto portfolio intelligence through ML-driven signals and backtesting on Crypt0nest. Replicated machine-learning-for-trading strategies across full modeling and backtesting pipelines in Python, trained gradient-boosted models on engineered momentum, volatility, and volume signals, and researched on-chain factors, building long-short portfolios from TVL-to-market-cap ratios.
Co-founded an AI dispatch startup and went through Y Combinator's Summer 2025 batch. Raised a $500K pre-seed round and scaled to $10K MRR within 3 weeks of launch, building the LLM voice-agent stack, data infrastructure, and evaluation harness that caught failures and improved outputs, then led pilots with real-estate firms to turn ambiguous market feedback into product decisions.
Helped build the playbook powering the next wave of agtech growth. Built a KPI dashboard and weekly reporting cadence for leadership and investors that surfaced trends, risks, and next steps, and analyzed 100+ farms to segment accounts and prioritize high-value opportunities.
Drove the data-driven future of how realtors close.
Finds the flight the airline won't offer after a cancellation: every airport in both metro areas, every carrier, live seats confirmed for your whole party, plus the DOT refund math on every option. Paste the airline's text and a language model fills the rest in. Built in Python on Duffel and OpenAI.
Prompt-first Roblox game creation. Describe a game in chat and an AI edits a real Roblox place: every change is a validated patch against a canonical instance tree, rendered live in a three.js preview, then built into a place file you can open in Studio and publish to your own account. Built with Next.js, TypeScript, and three.js.
A privacy-safe sponsored status line for the Claude Code terminal. No account, no code access, reversible in a single line. The wait while the model thinks becomes the placement.
Turns live NWS/NOAA hail data into the real homes inside a storm footprint, then scores and ranks them as roofing leads. Built with Next.js, Supabase, and OpenStreetMap.
A live algorithmic-trading and market-microstructure research lab for NBA prediction markets, built in Python on top of nba_api.
A wind tunnel for ideas. It takes an unfinished idea and forks it through a dozen context-isolated reasoning branches, each applying one creative operation like inverting the assumption or finding the cheapest disproof. Adversarial judges stress-test every branch, and a chairman synthesizes a decision packet, all streamed into one chat thread. Built with Node, Express, and the Claude API.
