Reference builds

Runnable, written-up seams around production agent systems. The point is not to show a demo; it is to make the architecture, evidence, and operating pattern concrete.

Published

• Agent memory over a lakehouse

  Typed, scoped, bi-temporal memory records over lakehouse data foundations; semantic retrieval as a derived index.

  Agentic AI
• Evaluation and observability harness for agent systems

  One typed run ledger projected into production trace, replay fixture, eval result, drift signal, and cost report.

  Agentic AI

Coming next

• Agent privacy as a data-flow architecture. PII classification, tool-call redaction, scoped memory writes, privacy boundary checks, and audit evidence.
• LLM gateway as the control plane for agent systems. Provider routing, fallback policy, cost budgets, latency SLOs, request logging, per-model eval gates, and redaction before provider calls.

Backlog

• Durable agent execution with checkpoints and cancellation.
• Enterprise RAG document extraction pipeline.
• Agent sandbox sized to blast radius.
• Verified-capability routing for multi-agent systems.
• Exception queue before happy-path automation.
• Policy-as-code for enterprise agent governance.