Privacy boundary
Raw community input is not shown publicly; the page uses synthetic sample text to demonstrate the publication path.
Pablo Zavala · AI Safety Evaluation · Research Engineering
Heard.now: Civic-Listening Platform
An early civic-listening pilot with Professor Jordan Usdan of Heinz College. It separates raw community input from published output: input enters an auditable store, while public extracts pass privacy and integrity checks.
Prototype architecture for an auditable input store and privacy-preserving public extracts
Public visuals use synthetic text so community messages stay private.
Role: Product and systems builder with Professor Jordan Usdan.
How to Inspect This Work
System pattern
The prototype separates private intake from public extracts, with verification checks before publication.
Scope
This is early-stage civic technology work with private materials, so the public artifact is intentionally a synthetic report card.
Case Study
Problem
Civic-listening tools can expose raw community input or publish summaries without a trustworthy audit trail.
Setup
Heard.now separates the private input store from public extracts, so publication can pass privacy and integrity checks.
Method
The prototype uses an auditable intake path, privacy checks, and verification checks before surfacing themes.
Result
The synthetic sample run shows privacy-preserving extracts and 7/7 verification checks without exposing raw messages.
Limitation
Public visuals use synthetic text because real community messages should not appear in a portfolio.
Evidence
The displayed report card is generated from a synthetic sample run of the local Heard workflow.
Key Outcomes
- Separates raw input storage from public extracts
- Uses synthetic public examples to avoid exposing community messages
- Shows a 7/7 verification check path in the sample run
Methods
- Privacy-preserving publication
- Auditability
- Synthetic test data
- Civic listening