AIHOOTS
// AI governance, built in the open

Proof, not slideware.

A public lab that builds enterprise-grade AI governance end-to-end — working, security-hardened, reproducible systems at the seam between generative AI and regulated environments. Every decision recorded. Every claim tested.

audit.jsonl — tamper-evident sha256 hash chain
seq 0ALLOWchat.completion requesta91f…c027
seq 1RESPresponse · 214ms · 87 tok4de8…1b90
seq 2REDACTpii: email removed77c2…9f5a
seq 3BLOCKprompt-injection patterne103…6d44
$ audit-verify audit.jsonl → OK: chain intact
The gap being filled

Everyone writes about AI governance. Almost no one ships it.

The space splits into prompt-tip influencers and slideware consultants. AIHOOTS occupies the empty middle: runnable reference systems that solve a real governance problem, built with the same engineering discipline the systems themselves demand.

01

Artifacts over opinions

Every episode is a system that runs, deploys, and can be rebuilt by a stranger from the published steps.

02

DevSecOps as the lesson

CI, SAST, dependency and secret scanning, SBOMs. Governed AI, built with governed engineering.

03

Honest metrics

Controls are measured, not asserted — precision, recall, and the misses are published alongside the wins.

Operating model

Build end-to-end first. Then publish so it can be retried.

Nothing is announced before it works — no "coming soon." Each system is self-contained, so the body of work compounds episode by episode instead of stalling in one unfinished platform.

BUILD

End to end, private

The whole system working locally against a containerized small language model.

HARDEN

Automate & secure

Pipeline green, security gates passing, architecture and decisions documented.

PROVE

Reproducibility

A clean-machine rebuild is executed and verified — if a reader can't retry it, it isn't done.

PUBLISH

Write-up + repo

The engineering story, the code, and the steps — public, together.

Episodes

Each one solves a single governance problem, completely.

E01 LLM Audit GatewayAn OpenAI-compatible proxy that produces a tamper-evident, independently verifiable audit trail of every request, response, and policy decision. ● live soon
E02 Guardrail & Injection Eval HarnessAn automated suite that scores an LLM endpoint against injection, jailbreak, and leakage cases — and runs as a CI gate. planned
E03 Governing an Autonomous AgentPutting a popular agent framework behind the gateway and measuring defense rates, with and without governance. planned