CivicLens

Landing Page
Urban Planning Agent, Map Simulation
Economist Agent
Equity Agent

Project Name

CivicLens

Project Category

Engineering and Infrastructure

Inspiration

In 2019, Berlin introduced strict rent control laws. Within 18 months, rental supply dropped by 56%. Ironically, the very people the policy aimed to help were worse off. Eventually, the policy was overturned-but only after years of damage.

That raised a simple but powerful question:

Every critical system in the world is tested before it goes live-planes use simulators, bridges are stress-tested, medicines go through trials.
So why are public policies deployed directly on millions of people without simulation?

It’s not because governments don’t care-it’s because tools like this don’t exist at an accessible scale.

We built CivicLens to change that.

What it does

CivicLens is a policy simulation platform-think of it as a flight simulator for government decisions.

A policymaker can:

Type or upload a policy in plain English
Instantly simulate its impact using real-world data
View outcomes through multiple perspectives
Explore results visually on an interactive 3D map

Behind the scenes:

We pull live economic and demographic data
Run three specialized AI agents (Economist, Urban Planner, Equity Analyst)
Combine their insights into a unified risk and impact score

The result?
A clear, data-driven preview of what might happen before the policy affects real lives.

How we built it

We designed CivicLens as a full-stack, multi-agent AI system:

Data Layer

US Census API -> demographic + housing data
FRED API -> macroeconomic indicators
Web search -> real-world policy outcomes

Intelligence Layer

Built a RAG pipeline using ChromaDB + sentence-transformers
Retrieved only the most relevant context per query to keep responses focused

Multi-Agent System

Economist -> market dynamics & incentives
Urban Planner -> spatial and infrastructure impact
Equity Analyst -> social and demographic effects

These agents run in parallel, each producing structured insights.

Orchestration

Used LangGraph to manage workflow:
1. Fetch context
2. Run agents
3. Combine results

Frontend

Built with Next.js, TypeScript, Tailwind CSS
Integrated Mapbox 3D visualization
Neighborhoods are rendered using:
- Height -> impact intensity
- Color -> risk level
Includes a time slider for long-term projections

Challenges we ran into

Census API returned HTML instead of JSON due to delayed key activation
Naming conflict (tavily.py) shadowed the actual library
NumPy version mismatch broke PyTorch dependencies
Mapbox race condition caused silent rendering failures
LLM outputs occasionally broke strict validation constraints

Each challenge pushed us to build more resilient and reliable systems.

Accomplishments that we're proud of

Built a working multi-agent AI system with real-time reasoning
Integrated live data, AI, and 3D visualization into one platform
Reduced response time from ~45s to ~15s using parallel execution
Designed a system that is both technically robust and user-friendly
Delivered a fast, reliable demo experience

What we learned

Multi-agent systems produce more balanced insights than single models
RAG is an architectural decision, not just a feature
Reliable demos matter more than complex ones
Most failures happen at system edges, not core logic
Visualization is key to making complex systems understandable

Impact

CivicLens enables policymakers to simulate outcomes before real-world implementation, shifting governance from guesswork to data-driven decision-making.
It helps identify unintended consequences, promotes more equitable policies by highlighting affected communities, and makes advanced policy analysis accessible to smaller governments.
Ultimately, it improves transparency, reduces risk, and supports better decisions that impact millions of lives.