VeriWrite - AI

Inspiration

Legal and policy teams waste hours manually scanning documents for contradictions. I wanted to automate compliance review using AI agents.

What it does

VeriWrite AI is an agent that detects contradictions in documents. It uses DeBERTa-v3 NLI to find conflicts like "delete after 30 days vs retain for 7 years" and logs every flag to MongoDB Atlas for legal audit trails.

How I built it

I built this with Google Cloud Agent Builder using Gemini 3 to orchestrate the workflow. The ContradictionDetector tool runs microsoft/deberta-v3-base for NLI. I integrated MongoDB MCP to write all contradictions to Atlas. Prototyped on Pydroid 3 for mobile development.

Challenges I faced

Running real NLI models on mobile was tough, so I simulated the DeBERTa-v3 output while preserving the exact Agent Builder + MongoDB MCP integration pattern for production deployment.

What I learned

How to chain Agent Builder tools with external MCPs like MongoDB for enterprise audit requirements.

What's next for VeriWrite AI

1. Deploy to production: Replace the simulated NLI with real microsoft/deberta-v3-base running on Vertex AI endpoints for live contradiction detection.

2. Google Docs add-on: Build a sidebar extension so legal teams can scan contracts directly inside Google Docs and flag conflicts in real-time.

3. Multi-document analysis: Expand the agent to compare contradictions across multiple PDFs and case files, not just single documents.

4. Enterprise dashboard: Add a MongoDB Atlas dashboard for compliance officers to search, filter, and export all flagged contradictions for audits.

5. Partner integrations: Connect to more MCPs like Confluence and Notion to scan internal wikis for policy conflicts.

Built With

• deberta-v3
• gemini-3
• google-cloud-agent-builder
• mongodb-mcp
• nli
• pydroid
• python
• vertex-ai