About
I’m Ruxi Zhang, a machine learning engineer interested in reliable AI
systems, agent execution, and ML infrastructure. lyr.ai is where I publish
experiments and the open-source systems I build around those questions.
The thread running through all of it: agents are stochastic systems, and most of what varies between two runs does not matter. Finding the part that does — early enough to act on it — is the problem I keep coming back to.
Writing
Experiments and research notes are on the front page, or by RSS.
Systems
- AgentSeism — measuring when execution variation becomes consequential, and testing whether agents can be steered before failure. State-level instrumentation, replay, fork, controlled intervention.
- TypedMem — schema-aware typed memory for AI agents: what an agent carries between steps, and what that does to its behavior.
- ReliAgent Bench — a reproducible reliability benchmark for memory-enabled agents; the harness whose failures drove the memory work.
- LYR — a layered knowledge engine.