Clausify AI

Clausify AI solves a real enterprise problem: procurement teams spend 4-6 hours manually cross-referencing contracts, invoices, and quotations — and still miss critical discrepancies that cost thousands. Upload your documents and Clausify instantly delivers: cross-document conflict detection (e.g., "invoice charges $525/unit but the contract says $425 — $13,144 overcharge"), risk analysis with severity ratings and source citations, side-by-side comparison matrices with clear winners, executive summaries written for decision-makers, and an AI chat copilot grounded in your documents with real-time streaming responses. The technical architecture runs 5 parallel LLM calls via asyncio.gather on AMD Instinct MI300X hardware through Fireworks AI, using a tiered model strategy: deepseek-v4-flash for deep reasoning and gpt-oss-120b (AMD-optimized) for structured extraction. The RAG pipeline uses 600-token chunks with 80-token overlap, all-MiniLM-L6-v2 embeddings, and ChromaDB vector search with top-12 retrieval. Key engineering highlights: consolidated conflict detection in O(1) LLM calls regardless of document count, 5-strategy JSON parsing for robust LLM output handling, persistent session storage surviving container restarts, auto-retry with exponential backoff, glassmorphism UI with full mobile responsiveness (320px-1920px), and professional PDF/DOCX report export. Built by Team Clausify: Rhenmart Dela Cruz ( Full Stuck AI Orchestrator Lead, Product Designer & ML ), Julie Ann Tiron (Backend & AI Research), Mica Pauline Calingo (Backend Support & Research), and Reymark Panes (QA Testing & Optimization).