Building data systems that see, stream, and scale.
Data engineer working on ML pipelines, real-time data platforms, AI agents, and developer tools.
Computer vision, streaming, and applied AI, with a soft spot for sports analytics.
|
Soccer video pipeline. YOLO finds players and the ball, a ByteTrack-style tracker follows them across frames, and the tracks become graphs of player movement for a GraphSAGE model. FastAPI serves batch and live video. MLflow is optional.
|
Soccer video experiments. One tool picks keyframes from a match video and has a vision model describe them. The other is an early player recognition package with no trained weights yet.
|
|
Multi-agent research system for biology. Ask about a protein and it pulls PDB structures, runs flexibility analysis, checks the literature, and suggests experiments.
|
🛠️ jod-skillsClaude Code plugins I use every day: a codebase audit that reads only the risky functions, a Hacker News digest, a GitHub trending digest, and an open-source status checker.
|
|
Real-time stock pipeline. Kafka streams ticks for 10 tickers, Airflow orchestrates, and a TimescaleDB star schema stores them. Runs with Docker Compose.
|
Flights, hotels, cars, deals, and billing as 13 services that talk over Kafka, run on Kubernetes, and sit behind a React client.
|
More: nobel-dataintelligence (protein stability from molecular vibrations) · contextbox (screenshot from the terminal, pull out text and links, search later) · voxt (voice-to-clipboard for Linux, in Go) · revolu-idea (multi-agent deep research)
I fix bugs in tools I use. Merged so far:
- lightpanda-io/browser: headless browser for AI. CSS selector spec fixes, Web API behavior, and CDP input events.
- Effect-TS/effect: TOML and YAML parsing edge cases, HTTP client and OTLP export fixes.
Also merged in radare2, SeaweedFS, uutils, Caddy, DSPy, the MCP Python SDK, and GitHub MCP Server.
| Languages | |
| ML & vision | |
| AI agents | |
| Data | |
| Infra & web |



