DenialShield

Denial Explanation
Claim Simulator
Pre Claim Analysis 2
Pre Claim Analysis 1
Appeal Letter

Inspiration

Insurance denials are confusing, time-consuming, and stressful. Patients receive rejection letters filled with codes and vague explanations, while doctors spend hours rewriting documentation and appeals. We were motivated by how unbalanced this process is: insurers use structured systems to deny claims, but patients are left without clear guidance on how to respond. DenialShield was inspired by the idea that if denials are driven by rules and patterns, then AI should be able to understand and challenge them.

What it does

DenialShield is an agentic AI assistant that helps patients and providers understand, prevent, and appeal insurance denials.

Users can upload medical bills, denial letters, EOBs, and doctor notes. The system: 1) Explains denial reasons in plain English 2) Identifies missing or weak documentation 3) Simulates how adding evidence improves approval chances 4) Generates professional, insurer-aware appeal letters 5) Learns from recurring denial patterns to improve future recommendations

How we built it

We built DenialShield as a full-stack GenAI system using PaddleOCR, open-source LLMs on Groq, and LangGraph for multi-agent orchestration. Different agents handle medical reasoning, policy analysis, legal framing, outcome simulation, and quality control.

To avoid a static system, we added a Denial Pattern Memory that tracks recurring denial reasons and successful fixes, allowing the system to improve without training or fine-tuning models. The frontend is built in React with a chatbot-style interface for document upload and interaction.

Challenges we ran into

Handling noisy real-world medical documents was difficult, especially ensuring reliable OCR and structured extraction. Designing prompts that were both deterministic and expressive was another challenge, particularly for appeal generation. We also had to be careful about balancing automation with explainability in a sensitive healthcare context.

Accomplishments that we're proud of

1) Building a fully agentic AI system using only free and open-source models 2) Implementing outcome simulation and counterfactual reasoning 3) Creating a learning system without model training 4) Delivering an end-to-end, realistic product rather than a static demo

What we learned

We learned that impactful GenAI systems are about orchestration, memory, and reasoning — not just model size. Clear explanations can be just as valuable as automation, especially in healthcare workflows.

What's next for DenialShield

Next, we plan to add a doctor-facing note-writing assistant, expand insurance coverage, integrate with EHR systems, and strengthen the learning loop using real user feedback.