Agentic Legal Risk Reasoning
Search standard names, typos, subsidiaries, and corporate aliases to instantly surface deeply nested legal conflicts across massive graph databases powered by Gemini 3 Flash.
During high-stakes legal operations, a firm must rigorously check for conflicts of interest before onboarding a new client or matter. Traditionally, these checks are manual, error-prone, and painfully slow—especially when corporate structures involve subsidiaries, shell companies, or complex naming aliases.
ConflictZero was born out of the need to eliminate this bottleneck. We leverage advanced graph databases, fuzzy matching algorithms, and state-of-the-art Agentic AI to transform a multi-day paralegal task into a millisecond computation.
ConflictZero provides an Executive Minimalist Search Dashboard where legal professionals can input:
- The Client's Name
- Opposing Parties
- Related Entities
In real-time, the system:
- Fuzzy-matches the inputs against thousands of records across organizations, contacts, and aliases.
- Traverses the Graph up to 3 hops deep to find indirect connections (e.g., Stark Industries → owns → Hammer Tech → opposing party in → active matter).
- Classifies the Risk intelligently:
- 🔴 High Risk: Direct match with an active matter.
- 🟡 Medium Risk: Indirect connections or fuzzy match typos.
- 🟢 Low Risk: No relational conflicts.
- Agentic Reasoning (Executive Counsel): Automatically structures the active graph payload into a context window and uses Gemini 3 Flash to deeply reason through the path hops, providing a deterministically actionable "Clearance", "Ethical Wall", or "Decline" formal risk assessment.
ConflictZero is built with a pure modern stack focusing on strict typing and high-speed data retrieval:
- Frontend: Built with Next.js, styled using Tailwind CSS, and utilizing highly customizable shadcn/ui components. We employed an "Executive Minimalist" light theme focusing on readability and professionalism.
- Backend: A rapid FastAPI Python service that orchestrates queries and manages the connection pool.
- AI Engine: Integrated via the new google-genai Python SDK utilizing
gemini-3-flash-previewon maximum reasoning thinking-levels to decode legal corporate hierarchies into exact mitigation strategies. - Database: Neo4j Graph Database. By leveraging nodes (Clients, Companies, Matters) and edges (Relationships), we drastically outperform traditional relational databases for recursive hierarchy lookups.
- Math/Algorithms: Cypher's APOC plugins are utilized to calculate the Sørensen–Dice similarity coefficient directly inside the database engine, avoiding costly data transfers to the application layer.
| Search Dashboard | High Risk Path Traversal |
|---|---|
 |
 |
| Fuzzy Match Detection | Expert Panel (Coming Soon) |
|---|---|
 |
 |
Everything is containerized for seamless development.
- Docker & Docker Compose
- Node.js > v20
docker-compose up -dDocker will boot Neo4j and the FastAPI service. Wait 15 seconds for Neo4j to be fully ready.
We have a golden seed script that deterministically creates the demo scenarios into the Neo4j instance.
# Ensure you are using your python virtual environment
cd backend
pip install -r requirements.txt
python seed_golden.pycd frontend
npm install
npm run devNavigate to http://localhost:3000. Wait for the "System Active" green pill in the top right, then search for Stark Industries to see a rich multi-hop High Risk conflict!
- Phase 6: Multi-language processing with Gemini 1.5 Pro to parse international legal briefs securely.
- Real-Time Webhooks: Integrating with Salesforce and common legal CRM platforms for automated night-time checks.
Built with ❤️ for this Hackathon.