ClaimGuard-Auditable AI Decision Engine

I didn't want to build another flashy demo. I came here to solve a real, painful enterprise problem: insurance claims automation. Manual processing is slow and vulnerable to fraud. A tired reviewer might miss a bad date on a form; an AI won't.My solution is ClaimGuard, a working prototype of an autonomous decision engine. The full stack is built (React, FastAPI), but the real win is the AI "brain" at its core.The Demo:When you upload a file, it triggers a 3-step AI pipeline:Gemini (Understand): The AI reads the claim data (using stable mock text in this demo to bypass local file-reading errors) and instantly extracts structured JSON.Qdrant (Compare): The system compares the data to historical fraud precedents. (This step is simulated in the demo, but a seed_qdrant.py script to power a live local Qdrant instance is included in the repo, proving the logic).Gemini (Decide): This is the magic. A second Gemini call analyzes all the data. In the demo, it instantly assigns a Risk Score of 90 (High Risk) and provides a perfect, auditable rationale:"The incident date of '2025-11-20' is a major inconsistency, occurring two days in the future..."It caught the fraud, and it explained why.How the Sponsor Tech Is Used:Google Gemini: Gemini is the engine's "brain." It's used for both extraction and the final, expert-level reasoning to score the risk. The entire "wow" moment of the demo is powered by Gemini.Qdrant: Qdrant is the system's "memory." Our architecture is built for it, and our seed_qdrant.py script proves the integration is ready for local deployment, bypassing the cloud quota errors we faced.Opus (AppliedAI): The Opus challenge is about the "Intake $\rightarrow$ Understand $\rightarrow$ Decide $\rightarrow$ Deliver" logic. My Python run_claim_analysis function is that auditable workflow. It's an Opus-ready logical blueprint, and I've included a workflow.yaml in the repo to show how it maps directly to the Opus canvas.This demo proves the core AI logic is sound.