SVP of Engineering at WP Engine. I lead a global engineering organization by day, and build open-source infrastructure for AI agents and physical AI on the side. 20+ years in distributed systems; Kraków → Austin.
I write about leading teams of people and agents at zygmuntdyras.com.
| Project | What it is |
|---|---|
| LOCI | Spatial memory for robots, drones and wearables: 4D (x, y, z, t) vector memory with novelty detection and an MCP server. pip install loci-stdb |
| LOCI Companion | Prototype: a chest-worn camera pendant that remembers what you saw, where and when. Build notes on my blog. |
| Enclave | A local agent that external AIs query over MCP, so they get answers instead of your documents. Early beta. |
| JEVTrader | A point-in-time research lab for scoring SEC 8-K filings with LLMs, labeling which backtest results are contaminated by model memory. Never places orders. |
| Hermes G2 HUD | Agent monitoring and control on Even Realities G2 smart glasses. |
- Low Context, High Stakes: Leading Engineering Across Four Cultures in the Age of AI
- Building LOCI Companion, Part 1: A Personal AI Assistant That Remembers What You See
- Agents Don't Need a Better Harness. They Need a Kernel.
- Context is the real work: when agents write the code, deciding what they should know is the scarce skill
- Agents need infrastructure: memory, permissions and scheduling
- Honest evaluation: labeling what can and can't count as evidence
- Local-first, privacy-preserving AI
Views are my own, not my employer's.