I build Python automation, AI-assisted tools, and production-ready data pipelines. I like systems that are fast, reliable, and easy to debug.
Selected projects
- Panopticon CVE Map
open-source— desktop app for vulnerability analysis with CISA KEV + local AI agent (MCP + OpenRouter). ~15 MB binary, ~80 MB idle RAM vs ~150 MB / ~300 MB Electron equivalents. dev.to. Code selected by XERJ Corpus Hub as a reference implementation for AI coding agents. - Murmur
open-source— Zero-Knowledge secret scanner for Git hooks (pre-commit + pre-push). SHA-256 hashes and masks only, raw secrets never touch disk, terminal or network. Hook overhead 29-80 ms on clean commits, Tauri 2 dashboard with 5-tab observability. v1.10.0. - Aero
open-source— Windows File Explorer opacity control with atomic config writes. - Market Analytics Bot
open-source— real-time arbitrage analytics with MCP server and Streamlit dashboard. - Sanity
private— high-frequency trading bot with WebSocket + REST fallback, 0 HTTP 429 on 56M+ items. Architecture documented in README. - GooEye
private— Goofish sniper bot with multi-user isolation, mtop API interception, 0 CAPTCHA on 44k+ items.
What I enjoy building
- Async systems that survive edge cases (WebSocket + REST fallback, retry with backoff)
- AI agents with structured tools (MCP servers, function calling, evals)
- Data pipelines that don't lose data (atomic writes, deduplication, cleanup)
- Desktop apps with local-first storage (SQLite, no cloud dependency)
- Security tools that respect privacy (Zero-Knowledge, no telemetry)
- Systems with contracts — clear inputs, predictable outputs, no magic
Maintaining and learning
- 🚀 Panopticon CVE Map — CISA KEV integration, MD export, versioned chat
- 🔒 Murmur — Zero-Knowledge secret scanner for Git hooks (pre-commit security layer)
- 📚 Learning: LangGraph, RAG architecture, eval frameworks (Langfuse/Ragas), ClickHouse