Full-stack and AI engineer
I build software end to end, and lately that means AI-native systems and the developer tools around them. Happiest when I'm deep in the codebase.
7+ years across the full stack, from native apps to developer tools other engineers use. I have led and hired engineering teams and owned architecture from the first conversation through shipped code. My deep specialty is commerce platform work (Shopify Plus, Hydrogen, custom Stripe gateways) for brands like JLO Beauty, Murad, and Hydrafacial.
What I want to build now is AI products and the infrastructure under them. The throughline across everything below is simple: use the right primitive for the job, and ship things people actually use.
An external memory for LLMs. Knowledge lives as a bi-temporal graph in Neo4j and gets retrieved with hybrid keyword plus vector search fused by reciprocal rank fusion, and the system learns from its own misses. Runs locally on Neo4j and Ollama.
A micrometaframework built on web primitives. Server-rendered HTML with client-side reactivity from native browser features (hidden inputs, labels, CSS :has()). No virtual DOM, no hydration, no bundler.
Mongo-style aggregation pipelines for Postgres. Write $match, $lookup, and $group, and ghostpres compiles them to a single chained-CTE query against jsonb while you keep ACID, foreign keys, and real indexes.
The Cursor experience for Neovim. Inline diffs, AI chat, and code context through the Claude CLI, with a Lua plugin talking to a Rust bridge.
A cross-platform fight-scoring app. One Svelte UI behind native iOS and Android shells plus an installable PWA, connected by a versioned bridge. Localized into English, Japanese, Russian, Portuguese, and Arabic, and fully offline-first.
Languages TypeScript · JavaScript · Python · Rust · Lua · SQL
Web & mobile React / Next · Svelte · Node · iOS / Android · PWAs
AI LLM integration (Anthropic, OpenAI) · RAG · vector search · MCP · agents · evals
Data & infra Postgres · Neo4j · Redis · AWS · Cloudflare · CI/CD