Work
Selected work
Seven public proof records across governed data preparation, grounded AI review, behavior evaluation, agent operations, visibility evidence, and ML product discipline.
-
systems proof Databricks CaseOps Lakehouse
Data platform / governed AI handoff
A public systems proof for governed document preparation before retrieval, AI review, or product claims. -
AWS Bedrock proof AWS Bedrock CaseOps Control Tower
Cloud AI operations / grounded review
A public AWS Bedrock proof for grounded document review with retrieval, validation, citations, structured outputs, and escalation boundaries. -
behavior evals lab Agent Behavior Evals Lab
Agent behavior evaluation / safety gate
A public lab for agent behavior evaluation: 320 real-agent records, a public leaderboard, and the Agent Behavior Safety Gate action, shipped publicly as Senthira. Its blind red-team audit caught 21.8 percent next to 98.0 on its own corpus; the low number stays published on purpose. -
operations lab Agent Terrarium
Multi-agent operations / open core
A public, sanitized extraction of a private multi-agent operations system. It ships a working console, stdlib-only tests numbered at 95 by its README, and an open-core section that names exactly what is held back. What is public here is the extraction, not the private system operating record. -
CI gate proof Agent Gate Demo
Continuous integration / unsupported agent claims
A minimal public demo where CI fails on a real pull request because the gate catches an agent claiming it ran tests it never ran. The red X on PR #1 is the artifact, and a visitor can verify it in under a minute. -
released prototype Evidence Loop Visibility Engine
Visibility evidence / reviewable proposals
A deterministic offline engine that turns visibility evidence into one reviewable proposal. Installable from PyPI, Apache-2.0, 4 tagged releases, green public checks. The test surface is small next to that release process, which is the gap to close first. -
ML product prototype E-commerce Purchase Intention MLOps
Data / ML product proof
A public prototype for purchase-intent prediction with reproducible evaluation, local serving, tests, model-card notes, and explicit production limits.