Claim Crusader

Inspiration

Medical claim denials cost U.S. healthcare providers $262 billion annually — and most go uncontested. Providers lack the staff and bandwidth to fight every denial, so revenue quietly bleeds out. We built ClaimCrusader to fix that.

What It Does

ClaimCrusader is an autonomous B2B appeal engine for healthcare providers. Drop in a denied EOB (Explanation of Benefits), and our multi-agent pipeline goes to work:

• Crusader Agent: Extracts the denial reason, CPT codes, and payer info, then cross-references a live knowledge graph of payer friction history to draft a targeted appeal letter.
• Critic Agent: Scores the draft (0–100), runs a compliance check, and flags anything that could tank the appeal before finalizing it.
• Assistant Agent: Helps with any operator questions to assist with appeal statistics and general inquiries.

The Self-Improving Loop

ClaimCrusader gets better over time. We utilize three core self-improvement strategies to ensure the system is constantly evolving its appeal logic:

• Knowledge Graph: Dynamically maps relationships between payers and specific denial codes.
• Knowledge Lake: An unstructured repository of policies, laws, and past patient appeals.
• LLM-as-a-Judge: An internal evaluation system that scores appeals and forces revisions before output.

Business Value

• Recapture denied revenue without adding headcount.
• Faster turnaround — appeals drafted in seconds, not days.
• Data-driven strategy — know which payers deny which codes most, and why.
• Scalable — one platform handles the full denial pipeline from intake to outcome tracking.

How We Built It

We built a multi-agent pipeline (Crusader → Critic) orchestrated via Airia, which acts as the enterprise API gateway and manages all of the agent's tool executions. The system is powered by Gemini for massive multimodal document extraction and deep-reasoning appeal drafting. We built an in-memory knowledge graph to track payer-level denial friction. Providers interact through a clean React frontend, while Lightdash plugs directly into our PostgreSQL database to track stats and win/loss metrics.

Challenges

• Designing a self-reinforcing feedback loop that meaningfully improves appeal quality over time.
• Building a robust, compliance-safe text pipeline under hackathon time constraints.
• Orchestrating multi-agent tool calls reliably across async backend flows.

Built With

• airia
• express-5
• gemini
• google
• lightdash
• lucide-react
• mcp-(model-context-protocol)-database-/-analytics-supabase-(postgresql)
• motion-v12
• nodemon
• react-19
• tailwind-css-v4
• ts-node
• typescript
• vite-6