Agent systems · memory infrastructure · developer reliability
I build and study infrastructure for long-running software agents: durable memory, inspectable workflows, and the reliability boundaries where agentic systems meet production software.
- Auditable memory and workflow continuity for long-running agents
- Reproducible experiments for agent behavior and developer tooling
- Evidence-driven fixes in open-source infrastructure
- memory-workbench — an experimental workbench for auditable memory and workflow continuity.
- Merged upstream contributions — evidence-driven fixes across agent infrastructure, databases, workflow orchestration, and developer tooling.
- Reproduce the real behavior before changing code.
- Prefer the smallest project-native fix with focused proof.
- Distinguish verified evidence from assumptions and unrun checks.
- Respect repository workflows, maintainer time, and contribution ownership.
I am especially interested in agent infrastructure, memory systems, developer tools, databases, and the small reliability bugs that become large at scale.