DocuSentinel AI

Inspiration Companies process thousands of invoices and contracts monthly, yet teams still manually open PDFs to compare amounts, track deadlines, and check missing documents. One missed mismatch can mean overpayments or compliance risks. We wanted AI that reasons across documents and proves every finding with evidence. What it does DocuSentinel AI classifies contracts, POs, and invoices, extracts key fields, cross-checks them, and flags anomalies, missing documents, and deadlines — each linked back to the exact source page in the original PDF. How we built it A 5-stage pipeline: Ingest → Classify → Extract (OCR + LLM with page coordinates) → Reason (cross-document rules) → Present (findings + evidence). We also built a full interactive web simulation and a professional 6-slide pitch deck. Challenges we ran into Making multi-document reasoning transparent, designing trustworthy evidence links, and explaining complex logic simply for non-technical audiences. Accomplishments that we're proud of A clear end-to-end simulation, severity-coded findings with one-click evidence, and a judge-ready presentation that turns a complex problem into a 2-minute story. What we learned Extraction alone is not enough — cross-document reasoning and provenance matter most. A strong real-world example (₹70k mismatch + missing certificate) communicates value faster than architecture diagrams. What's next for DocuSentinel AI Real OCR/LLM integration, more document types, ERP connectors, human-in-the-loop review, on-premise options, and broader anomaly detection.