About the Project Inspiration

Accessing government policies, social welfare rules, and public programs is often confusing, inaccessible, and full of legal jargon. Many people simply don’t apply because they can’t understand the eligibility or process. We were inspired to build something that translates bureaucracy into conversation — so every citizen can understand and act on their rights.

What it Does

You upload a policy or ordinance (PDF, document text).

You ask questions in plain language (e.g. “Am I eligible for housing assistance?”).

The system parses and reasons through the document, extracts relevant rules, and gives you a personalized answer with steps.

You can also view the AI’s reasoning trace — how it arrived at that answer by referencing sections of the document.

How We Built It

Document Ingestion & Embeddings: We chunk text, create embeddings (vector store) to allow semantic search.

Agent Architecture:

Parser Agent — splits and understands policy text

Eligibility Agent — reasons about user context + rules

Explainer Agent — translates logic into conversational output

Frontend: React + Tailwind + animated UI components

API Layer: Backend endpoints coordinate the agents and data

Deployment: Hosted on Vercel / serverless backend

In development, we used prompt engineering loops and test documents to refine each agent’s accuracy.

Challenges We Ran Into

Getting AI to avoid hallucination when documents are ambiguous

Handling scanned or low-quality PDFs and extracting structured text

Dealing with conflicting rules across multiple policies

Managing user context in conversation (keeping state, asking clarifiers)

Performance — latency when reasoning over large documents

Accomplishments We’re Proud Of

Built a working MVP in under 48 hours

Successfully parsed real policy documents and answered user queries

Designed a reasoning trace visualization to provide auditability

Demonstrated agentic AI collaboration — humans prompt, AI designs workflows

Created a UI experience that feels credible, responsive, and alive

What We Learned

How to chain AI agents effectively (parser → reasoner → explainer)

Techniques for semantic retrieval + prompt refinement

UX lessons in transparency: users want to see why the AI gives an answer

The importance of error-handling and fallback flows (when AI isn’t confident)

What’s Next for gov-clarity-bot

Multilingual support — bringing it to multiple languages

Voice / multimodal input/output (ask by voice, get spoken response)

Expand domain coverage (local governments, municipal policies, taxes)

Collaborations with civic groups or governments to adopt it as a public tool

Add continuous document ingestion — daily auto-updates when new policies are published

Built With

Share this project:

Updates