Core Engineer at Bread, building the rails for the agent-native web.
- OpenLens — an AI-visibility tool designed to help you observe, control, and take action on the agents already impacting your day-to-day operations.
- OpenLens CLI — a user-personalized agent harness in Go, with real OAuth and a chat-first interface.
- Agent Honeypot - a universal WebMCP agent harness that allows AI agents to explore the web, freely, without restriction.
Open weights, open data, honest evals. Work on small specialist models that punch above their size:
- GLiNER distillation / fine-tuning — teaching compact NER models to match much larger teachers on extraction, with recall that clears the bar at a fraction of the cost.
- Reproducible pipelines: distill, quantize, evaluate, publish — so the next person starts where I finished, not from zero.
🔬 Earlier — @vitalune
Research and tooling that seeded a lot of the above — model surgery, constant iteration + deployment lifecycles, and the conviction that neural networks can erase the limits we've always called human.
- System design is everything. trading PhD-depth theory for product and system-design intuition allow me to ship at inference speeds.
- Democratization. If it only works for a lab with a cluster, it doesn't count. If it can be restricted to any human beings, it doesn't count.