.png&w=256&q=75)
Yashas Nagesh Gowda@YashasGowda
1
Events attended
1
Submissions made
Looking for experience!
Socials
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
Shield AI — Enterprise Contract Intelligence
Enterprise legal teams review hundreds of contracts every month — NDAs, vendor agreements, employment contracts, data processing addenda — each requiring expert analysis for risk, compliance violations, and adversarial content. This process is slow, expensive, and error-prone. Shield AI solves this with a production-grade, AI-powered governance platform built on a 6-agent pipeline powered by Google Gemini 2.5 Flash-Lite. Agent 4, the Security Gate, runs first — before any LLM sees the document. Using Veea's Lobster Trap DPI proxy and an offline pattern detector, it scans every upload for prompt injection attacks, hidden text, and policy bypass attempts. Threats are quarantined instantly. The LLM never sees them. Agent 0 classifies the contract type, jurisdiction, and applicable frameworks. Agent 1 extracts all structured data — parties, dates, obligations, and clauses. Agent 2 performs semantic risk scoring using Pinecone RAG, producing a composite score from 0–100. Agent 3 runs compliance verification against HIPAA, GDPR, SOC 2, CCPA, and PCI-DSS per clause. Agent 5, a pure policy engine with no LLM, deterministically routes every contract to AUTO-APPROVE, MANAGER-REVIEW, LEGAL-REVIEW, or REJECT. The platform features role-based review queues for five roles — Procurement Analyst, Compliance Officer, Legal Reviewer, Executive, and Auditor — with mandatory decision reasoning that feeds back into future LLM outputs. A natural language interface lets users query any contract in plain English. An immutable audit log captures every agent run, human decision, and security event. A live scoring policy editor lets administrators adjust weights and rescore the entire corpus in real time. Low-risk contracts auto-approve in under 15 seconds. High-risk contracts reach the right reviewer with full AI rationale and clause-level findings. Adversarial contracts are blocked before any model processes them.
19 May 2026
