Building tools that extend what AI can understand, remember, create, and accomplish.
I explore the systems around AI agents - context, memory, knowledge, tools, workflows, verification - so they become capable collaborators instead of confident guessers.
🧠 Understand - structured, grounded knowledge instead of whatever's in the weights.
RAG provenance structured data
🗂️ Remember - context that survives the session, the project, the agent.
persistent context state Git-backed workflows
🤖 Act - the tooling that lets an agent actually get things done.
orchestration MCP automation
🔍 Verify - measure what happened, don't trust the summary.
validation observability deterministic checks
cdx-manager - orchestration for AI coding agents: isolated environments, sessions, quotas, context handoffs, machine-readable output.
One workflow → Multiple agents → Shared context
logics-manager - persistent context and structured workflows for AI-assisted development. A local-first runtime turning ideas into traceable work.
Context → Plan → Task → Implementation → Verification
electrical-plan-editor - local-first workspace for designing, validating and exporting electrical plans.
Complex domain → Structured model → Validation → Output
meshanvil - lets AI operate on 3D applications while producing measurable evidence of what actually happened.
Operate → Measure → Prove
deepvault - governed knowledge sources for AI, with provenance, permissions and evidence preserved.
Knowledge → Retrieval → Reasoning → Evidence
day-captain - Microsoft 365 mail and calendar turned into actionable daily intelligence.
Deterministic where it matters → Intelligent where it helps
Building the tools that make AI better at building things.