Anthropic x Penn Hackathon — Champion Track, 3rd Overall

Heard

A democratic advocacy platform that converts citizen grievances into ready-to-send policy briefs, maps elected officials across 4 government levels, and pairs one-click citizen outreach with a district-aware official dashboard.

Civic Tech Multi-Agent Full-Stack Geospatial

The Problem

Citizens have complaints about local infrastructure, public safety, and policy failures, but turning frustration into effective advocacy is hard. Most people don't know which official to contact, what jurisdiction the issue falls under, or how to frame a complaint that gets attention.

Heard closes this gap with a two-sided platform: citizens describe a problem in plain language and receive a complete advocacy toolkit — policy brief, root-cause analysis, affected population estimate, and a ready-to-send letter addressed to the right official. On the other side, elected officials see a district-aware dashboard that aggregates constituent grievances, tracks trending issues, and surfaces directed actions by jurisdiction.

The platform maps officials across municipal, county, state, and federal levels using the Cicero API, with privacy-preserving geolocation that drifts display coordinates by ~200m while storing exact locations privately.

Technical Highlights

  • 4-Level Official Mapping

    Automatic district assignment via Cicero API spanning municipal, county, state (house + senate), and federal levels, including at-large seat handling.

  • AI-Powered Advocacy Pipeline

    A civic intelligence skill generates policy briefs, root-cause analysis, intervention recommendations, and professional letters from unstructured citizen complaints.

  • Two-Sided Platform

    Citizens get one-click outreach; officials get a dashboard filtered by their exact jurisdiction showing grievances, trending issues, and directed actions with engagement stats.

  • Community Engagement

    Public grievance feed with full-text search, follow tracking, and trending issue aggregation. Privacy-preserving display locations on an interactive map.

Under The Hood

Agent Orchestration

Built on DeerFlow 2.0 (ByteDance) with a LangGraph server for multi-agent orchestration. The civic intelligence skill handles the full pipeline from complaint parsing to letter generation, with sandboxed execution and MCP integration for extensibility.

Geospatial Backend

PostgreSQL with full-text search and geospatial indexing. District-level indexes enable fast filtering for the official dashboard. The Cicero service caches official lookups for 24 hours and extracts district assignments across 8 jurisdiction types.

Frontend

Next.js 16 with React 19 serving a chat interface for grievance input (voice and text), a community feed with map visualization via MapLibre, and a candidate dashboard with district-aware filtering.