Building products for a global audience is tricky. Every new feature might be perfectly fine in one place but raise red flags elsewhere. Think California SB-976, the EU DSA, GDPR, the list keeps growing.

Today, compliance checks are:

• Slow and costly — lawyers comb through every feature
• Inconsistent — different reviewers → different verdicts
• Risky — laws evolve quickly, so things get missed

We wanted a way to catch geo-regulation issues early — with evidence that’s clear, auditable, and fast enough to keep up with product cycles.

We built JurAI, a multi-agent AI pipeline that works like a courtroom for compliance.
Given a feature description and a target region, JurAI delivers a JSON compliance report that spells out:

• Whether region-specific compliance is required
• Which clauses and laws apply
• Why the decision was made (with sources you can trace)

Instead of guesswork, you get an auditable verdict you can plug directly into legal review or CI/CD.

1. Dual Retrieval

• Past Verdicts: broad coverage using vector similarity
• Legislation: precise matches using entity-based search

2. Jury & Critic System

• Multiple Jury Agents analyze the feature independently
• Each Jury has its own Critic Agent that reviews, points out weaknesses, and forces revisions
• A Judge Agent merges jury outputs, removes duplicates, and delivers one clean verdict

3. Diversity of Models

• Juries don’t all run on the same LLM — e.g., one might use DeepSeek, another GPT-5-mini
• Critics and Judge can mix models too
• This way, no single model’s blind spots dominate the outcome

4. Structured, Human-Friendly Output

• Always JSON, never free-text rambling
• Every statement backed by a citation
• Easy for lawyers and pipelines to parse

• ⚡ Faster reviews — first-pass compliance in minutes, not weeks
• 🔍 Transparent — every verdict comes with references
• 🧩 Modular — drop in new laws/regions as needed
• 🧠 Robust — multiple models + self-critics → fewer blind spots
• 💡 Engaging — it feels like watching AI lawyers argue and a judge decide

.
├── frontend/ # Next.js UI (feature input + report display)
├── backend/ # Google ADK agents (Jury, Critics, Judge, RAG tools)
└── README.md # This file

Each subfolder has its own README with setup instructions.

• Expand to more regions (APAC, LATAM, etc.)
• Benchmark against human lawyer reviews
• Batch processing for cheaper API calls
• Feedback loop so users can refine verdicts
• Let users pick which LLMs power their juries/critics

cd backend

# create and activate venv
python3 -m venv .venv
source .venv/bin/activate

# install dependencies
pip install -r requirements.txt

# run server
uvicorn main:app --host 0.0.0.0 --port 8000 --reload

cd frontend

# install dependencies
npm install

# run dev server
npm run dev

Create a .env.local file inside frontend/:

NEXT_PUBLIC_API_BASE=http://localhost:8000

This tells the frontend where to find the backend API.