Note: This repository serves as a technical showcase for a proprietary application. Due to NDA and IP restrictions, the source code is not public.
Showcasing a specialized, internal AI system designed to automate the analysis and comparison of complex business documents (quotes, contracts, NDAs). Acting like a domain expert, this platform will validate incoming documents against:
- Industry Standard Conditions (Sector-specific legal frameworks).
- Internal Compliance Checklists (Client-specific business rules).
- Commercial Risk Logic (Price validity, liability caps, etc.).
During the first month of deployment, the system demonstrated immediate value:
- Concrete Discrepancies: Identified 29 specific mismatches between Purchase Orders (POs) and Sales Contracts.
- Critical Risk Mitigation: Flagged 7 critical and high-impact risks related to liability clauses, compliance gaps, and conflicting Terms & Conditions.
- Operational Protection: Directly prevented financial and operational exposure by proactively detecting subtle deviations in delivery schedules, payment terms, and scope creep.
The system utilizes a multi-agent architecture where a central orchestrator routes user requests to specialized agents:
- PDF Tool: Handles OCR, text extraction, and general document comparison.
- Compliance Validator: Specifically checks documents against strict industry terms.
- Risk Assessor (KPI Agent): Structures the output into risk categories and assigns a Severity Score (Low to Crucial).
- Context-Rich Analysis: The LLM is injected with specific business context, preventing generic hallucinations.
- Severity Matching Algorithm: Distinguishes between trivial text differences (e.g., phrasing) and critical contractual risks (e.g., liability terms).
- Multi-Agent Orchestration: Powered by n8n, allowing for complex, branching logic flows that single-prompt LLMs cannot handle.
- Secure Authentication: Role-based access control (RBAC) via Supabase Auth with manual admin approval flows.
- Real-time Dashboard: Visualizes risk distribution and historical analysis using Recharts.
React, TypeScript, Vite, Tailwind CSS, shadcn/ui, Supabase, PostgreSQL, n8n, Anthropic Claude, Google Gemini.
Developed by Jack van der Vall in collaboration with a Dutch industrial piping and prefabrication contractor specializing in large-scale energy and utility infrastructure.
Copyright © 2025. All rights reserved.