Work

An AI analyst I run myself: every hour it reads 127 sources, runs a nine-stage LLM pipeline with a fallback model behind every stage, and writes one page of market conclusions that trace back to their posts.

• 9LLM stages a run
• 127sources an hour
• 4,100+reports written

• Next.js 16
• React 19
• TypeScript
• Vercel AI SDK v6
• SQLite FTS5
• Live demo

Graduate capstone on serving latency: I owned the KV-cache layer under a learned vLLM scheduler, where 4-bit KV fit 2.83 times the tokens and cut time to first token 6.8 times under load.

• 2.83×token capacity
• 6.8×faster first token
• −47%P99 time per token

• Python
• vLLM
• Triton
• PyTorch
• Llama-3.1-8B

A bundle relay built for 30,000 to 60,000 requests a second across eight block-engine regions, and the market research that decided what to build on top of it.

• 4,000+egress IPs
• 8block-engine regions
• <0.5 msproxy wait, from 13–28

• Go
• gRPC
• Rust (tokio, axum)
• Solana Go SDK
• Jito

A Rust trading system for Polymarket's 5- and 15-minute BTC Up/Down markets: it reads Binance over SBE ahead of the market's own feeds, prices each market as a binary option, and trades only when the book sits far enough from fair to pay for the fee and the hedge.

• 720 / 894commits mine
• ~1.65 sahead of the settlement feed
• 17,549markets in the backtest corpus

• Rust (tokio)
• Binance SBE
• Polymarket CLOB
• Gnosis Safe, CTF
• Python

I co-founded a crypto asset database and designed and collected its data from zero; it grew to 428K users and 200+ API partners, and I later opened the data to AI agents through MCP.

• 428Kusers
• 200+API partners
• ~14×organic growth in a year

• Product management
• MCP (TypeScript, Python)
• Cloudflare Workers
• XGBoost
• Streamlit
• Live demo

Rebuilds stablecoin card-payment volume from self-indexed chain data, issuer by issuer, with an evidence trail behind each figure.

• 22 of 24card programs traced
• 413registered addresses
• ±0.6%on EVM mainnets

• TypeScript
• ClickHouse
• Postgres
• EVM adapters
• Solana adapters

A bilingual plan catalogue for a telecom reseller: staff edit plans in an embedded studio and the site updates without a redeploy. The live demo runs on fictional data.

• 5product lines
• 2languages
• 9content schemas

• Next.js App Router
• TypeScript
• Sanity v3
• GROQ
• next-intl
• Live demo