← Archive

Foundation · Public sector programme · NDA

Public data chatbot with an audit trail first

A national statistics agency cannot publish answers nobody can trace. Fluency without an audit story is a liability, not a launch.

Lead outcome

Hours → sec

query response we had to beat

Focus

Conversational AIPublic SectorNDA
StackLLMsVector DBSpeech AI

What I owned

I insisted the audit and explainability path ship before conversational polish — four census cycles of source data, zero appetite for untraceable answers.

Narrative

Under NDA for a national statistics context. The model work was necessary. The programme work was making explainability a gate, not a footnote.

My seat: sequencing audit trail, source citation, and stakeholder sign-off so a public-sector buyer could defend every answer.

Same pattern I run elsewhere — absorption and accountability before demo polish.

Audit story before fluency

If you cannot show why an answer exists, it does not ship — regardless of latency.

Public trust is the definition of done

Seconds matter only after traceability clears the gate.

Hours → sec

query response we had to beat

Gov-grade

traceability bar I held

4 cycles

census source depth in scope

Next step

Want the confidential layer — names, constraints, steering artefacts? That conversation stays over email.

Email me about this case