Nripanka Das — AI product leader building inspectable agent infrastructure and decision systems for energy, climate, and product outcomes.

• Trustline MCP — decide whether a tool call is allowed, denied, or held for human approval. Source · v0.1.1
• Grid Ops Arena — replay a seeded microgrid outage and compare reliability, cost, emissions, and safety. Source · v0.1.1
• Value Density Lab — compare product outcomes against time, friction, regret, and harm. Source · v0.1.1

Each report is generated by a deterministic offline demo. No account, API key, employer data, or silent telemetry is required.

• Specify and test — SpecSpan links requirements to code and tests; ToolDrill attacks tool-contract boundaries.
• Inventory and control — Agent SBOM surfaces capabilities; Trustline MCP enforces policy, approval, quota, and audit rules.
• Replay and govern — RunMirror records deterministic cassettes; Memory Gauntlet tests correction, deletion, expiry, and role isolation.
• Exercise real decision shapes — Grid Ops Arena, Carbon Risk Lab, Climate Evidence Bench, and Value Density Lab.
• Leave reviewable evidence — TraceWeave, SandboxLedger, and ProofDeck turn runs into inspectable records.

• PatchGym turns Git history into a local coding-agent benchmark with hidden tests and oracle patches.
• SpecForge is a live, local-first studio for turning research evidence into inspectable project blueprints.
• ProofDeck creates static evidence bundles whose contents and Merkle root can be independently checked.

The August 16, 2026 release gate verified 10 public v0.1.1 releases, 1,020 source and extracted-package test executions, 26 live release assets, three current-main proof sites, and zero open CodeQL, Dependabot, or secret-scanning alerts at audit time.

• Post-release stress and security audit
• Reliable AI release architecture
• Deep profile and landscape audit

My working rules are simple: high-stakes automation needs human takeover; evaluation must leave evidence; composite scores must expose their parts; and benchmarks must state their limitations.

I work across AI product leadership, energy, and supply chain. This is an independent, code-first lab with no employer branding or implied endorsement. Professional context and writing live on LinkedIn.

Found a reproducible bug or a useful integration? Open an issue in the relevant repository. Specific technical criticism is welcome.