Public Engagement Coach

A coach that turns a rough idea into a public engagement plan.

9Framework questions

Year: 2026
Role: Technical lead
With: InnovateUS and The GovLab
Category: Government AI

01/The problem

The problem

Practitioners who run public engagement projects rarely have an expert to ask. The GovLab framework is good, but it is a PDF.

02/What I built

What I built

01
A chat that walks people through the nine-question framework, with an orchestrator that routes each message to a coaching, retrieval or suggestion agent.
02
Answers draw on a curated case-study library and anonymized expert interviews in Weaviate, and end with an evidence-grounded reflection.

03/Architecture

How it flows

Each message is classified, sent to the matching agent, grounded by hybrid search, and streamed back.

04/Outcomes

What it did

9
Framework questions
The GovLab nine-question framework the coach is built around.
From the repo

Check these in the ledger

05/Production notes

Behind the numbers

All receipts for this project

Mechanism

An orchestrator classifies every message into one of five actions (open a coaching question, continue it, retrieve examples, suggest the next question, or general) and sends it to a prompt-backed agent. Answers cite case studies from a Weaviate library through hybrid search.

Guardrails

A Pro Tip from the anonymized expert interviews appears only after a gatekeeper and a judge pass it, and never blocks the answer. Interviews are redacted with Comprehend, and the interviewer’s name never enters a model, Step Functions or Weaviate payload. Search returns nothing instead of weak matches.

Evaluation

Not measured

No evaluation of the routing or the Pro Tip judge is in the repo.

Evidence

Private repo

The repo is private. I was technical lead; teammates wrote most of the code.

06/Stack