VendorLens

🎯 Inspiration

Enterprise vendor onboarding is broken. When Goldman Sachs evaluates a new SaaS vendor, analysts spend weeks manually researching security certifications, compliance policies, API documentation, pricing models, and support capabilities. This process involves dozens of spreadsheets, countless browser tabs, and mountains of PDFs—all to answer one question: Should we trust this vendor?

We saw an opportunity to transform this tedious process into an intelligent, transparent AI workflow that doesn't just automate research—it shows its work in real-time.

💡 What It Does

VendorLens is a multi-agent AI system that autonomously evaluates vendors across five critical dimensions:

The 7-Agent Pipeline

Intake Agent - Normalizes vendor data and extracts key information
Verification Agent - Fact-checks claims against official sources
Compliance Agent - Discovers and analyzes security certifications (SOC2, ISO27001, GDPR)
Interoperability Agent - Evaluates APIs, SSO support, and integration complexity
Finance Agent - Estimates Total Cost of Ownership (TCO) from pricing documentation
Adoption Agent - Assesses support quality, training resources, and implementation timelines
Summary Agent - Synthesizes findings into weighted recommendations

Two Powerful Workflows

📋 Vendor Application - Single vendor applies for onboarding, receives comprehensive evaluation

⚖️ Vendor Assessment - Compare multiple vendors side-by-side with customizable priority weights (security vs. cost vs. integrations)

🔴 Real-Time Transparency (Our Secret Weapon)

Unlike black-box AI systems, VendorLens streams every agent's reasoning live via Server-Sent Events (SSE). Watch as agents:

🔍 Discover documentation URLs using LLM-powered web search
💭 Analyze content with Nemotron and extract structured findings
📊 Calculate scores with confidence levels based on source credibility
✅ Pass context to the next agent in the pipeline

The frontend displays a live workflow visualization showing which agent is active, what it's thinking, and what it discovers—transforming opaque AI into a transparent, trustworthy research assistant.

🛠️ How We Built It

Tech Stack

Frontend: Next.js 14 (App Router), React, TypeScript, Tailwind CSS, Recharts
Backend: Python, FastAPI, MongoDB
AI: NVIDIA Nemotron via NIM (local deployment to bypass rate limits)
Real-Time: Server-Sent Events (SSE) for live agent streaming
Document Processing: PyPDF2 for extraction, custom RAG retrieval