PolicyForge: AI Automation Policy Simulator

Autonomous agent that continuously monitors AI regulations worldwide, simulates economic impacts of policy changes, and generates actionable implementation roadmaps for governments.

Comment

PolicyForge AI

A visual regulatory intelligence platform that transforms how organisations navigate the global AI compliance landscape.

💡 Inspiration

Governments worldwide are racing to regulate AI, creating a fragmented patchwork of conflicting rules. The EU AI Act, US Executive Orders, and emerging policies across Asia all operate in silos - leaving organisations struggling to understand their compliance obligations.

I wanted to move beyond simple "chat with PDF" interfaces. PolicyForge was born from the idea that an AI analyst should be visual, structured, and context-aware. It leverages Gemini 3's "Thinking" capabilities to reason through regulatory complexity and presents results not as walls of text, but as interactive maps, Gantt charts, and comparative tables.

🎯 What It Does

PolicyForge is a comprehensive regulatory intelligence dashboard with five distinct analysis modes:

Mode Description
Monitoring A live feed of regulatory changes, filtered by region, using Google Search grounding for up-to-the-minute accuracy
Simulation Models specific regulation impacts on different entity types (e.g., "How does the EU AI Act affect a HealthTech startup?") with sector-specific risk assessments
Comparison Generates side-by-side matrices of jurisdictions (e.g., US vs. EU), highlighting key divergences in liability and enforcement
Roadmap Automatically generates implementation timelines with phases, resource requirements, and dependencies - visualised as interactive Gantt charts
Trends Analyses policy velocity (speed of legislation) and identifies emerging themes with detailed explanations and news citations

The app features an interactive global map allowing users to drill down from global views to specific states (like California or Bavaria) to set the context for the AI's analysis.

🛠️ How I Built It

PolicyForge is a React application powered entirely by the Google GenAI SDK.

Architecture & AI Strategy

Hybrid Model Approach

Model Use Case
Gemini 3 Pro Preview Heavy-duty reasoning tasks (Simulation, Comparison) - utilises Thinking Config (Chain of Thought) to plan analysis steps before outputting JSON
Gemini 3 Flash Preview Low-latency tasks - Dashboard updates, Trend analysis, and Autosuggest features

Key Technical Decisions

  • Strict Structured Output - The application relies on complex JSON Schemas. The AI returns strictly typed data structures for charts, timelines, and comparison tables, rendered using Recharts and custom components
  • Real-Time Grounding - The googleSearch tool is integrated into the analysis pipeline, forcing the model to cite real-world sources and URLs for every claim
  • Interactive Mapping - Built with D3.js and TopoJSON, handling dynamic zooming and state-level selection, feeding geographical context directly into system prompts

🧗 Challenges I Faced

Taming the JSON Schema

Getting an LLM to consistently output deeply nested JSON arrays - specifically for Comparative Analysis tables and Roadmap Gantt charts - was difficult. I had to refine schema definitions rigorously to ensure the UI never crashed due to malformed data.

Visualising "Thinking"

Gemini 3 Pro with a high thinking budget provides excellent results but takes time to process. To improve UX, I built a custom Thinking Indicator that simulates the agent's internal monologue (e.g., "Scanning legal frameworks...", "Calculating economic impact...") so users understand the agent is actively working, not just loading.

Map Data Granularity

Mapping global jurisdictions down to state/province level (like specific German states or UAE Emirates) required handling complex TopoJSON overlays and coordinate systems to ensure smooth zooming transitions.

📚 What I Learned

Chain of Thought is a UI feature - The "Thinking" process isn't just an internal backend mechanic. Exposing the Reasoning Trace to users builds trust. Users love seeing how the AI reached a conclusion.

Context is King - Passing specific map selections (e.g., "Selected: California") into the prompt dramatically improves output quality compared to generic queries.

Structured UI > Chat - For complex domains like law, a dashboard with charts and tables is infinitely more valuable than a linear chat interface.

🚀 What's Next for PolicyForge

Feature Description
Multi-Modal Inputs Allow users to upload specific PDF drafts of bills to simulate impacts before they're even passed
Export Capabilities Generate PDF reports of Roadmaps and Comparison tables for policymakers to download
User Personalisation Save simulation profiles (e.g., "My SaaS Company") so the agent remembers user constraints across sessions

🔧 Built With

  • React
  • Google GenAI SDK
  • Gemini 3 Pro Preview
  • Gemini 3 Flash Preview
  • D3.js
  • TopoJSON
  • Recharts

Built With

Share this project:

Updates

posted an update —

Dogfooding PolicyForge: UK vs EU AI Regulation

I wanted to test whether PolicyForge could answer a question I genuinely care about professionally: how does the UK's voluntary, principles-led approach to AI regulation compare against the EU AI Act's mandatory framework? I selected the EU and UK on the interactive map and ran a comparative analysis using Gemini 3 Pro's thinking mode.

Selecting EU and UK jurisdictions on the interactive map

PolicyForge thinking indicator showing reasoning steps

The structured comparison table surfaced dimensions I'd have had to pull from multiple sources manually - like the EU's high-risk system guidance being delayed into February 2026, or the UK's reliance on the Data (Use and Access) Act 2025 as a stopgap while its formal AI Bill slips to late 2026. It also correctly identified common ground around GPAI safety and regulatory sandboxes, all grounded with real cited sources rather than training data.

Comparison results with structured table and confidence score

This is exactly the kind of workflow I built PolicyForge for: turning a complex, multi-jurisdiction regulatory question into a structured, citable output rather than a wall of text you have to parse yourself.

Log in or sign up for Devpost to join the conversation.