Inspiration
We were inspired by the overwhelming number of people who, despite having health insurance, face unexpected medical denials. These rejections leave patients feeling helpless and burdened with massive medical debt for their own treatments. In reality, insurers often reject claims for opaque reasons that prioritize their own profit margins over patient care. AppealForge was born to empower these individuals, giving them an accessible tool to fight back and claim the financial support they are rightfully paying for. What it does
AppealForge is an AI-powered clinical appeals system. Through a FastAPI backend, the system ingests denial letter PDFs and uses AI to analyze the content, extract critical clinical codes (like CPT and ICD-10), and gather context. It then generates a robust, legally sound appeal draft. Through an interactive UI, the user can verify the information, modify the text, and easily export the final appeal as a Markdown or PDF document ready for submission. How we built it
We built a complex data pipeline in just 48 hours using a modern stack: Python and FastAPI for the backend, and TypeScript with React for the frontend. To power the intelligence, we implemented a Dual-Model Architecture using LLMs provided by Featherless AI (DeepSeek-V4-Pro for drafting and Qwen 2.5 32B for clinical auditing). By combining our diverse skill sets, we successfully connected these technologies into an efficient, RAG-augmented workflow that materialized our vision into a highly functional tool. Challenges we ran into
Technically, implementing cutting-edge AI features—like ensuring deterministic RAG grounding and preventing LLM hallucinations—within a strict time limit pushed our technical boundaries. On the collaborative side, moving fast under a 48-hour deadline led to constant Git merge conflicts. Learning to navigate version control safely while maintaining our development speed was one of our biggest hurdles. Accomplishments that we're proud of
We are incredibly proud of successfully structuring a complex, multi-layered software architecture that we initially thought was beyond our current capabilities. Furthermore, mastering asynchronous collaboration as a four-person team and delivering a polished, working product from scratch is a huge milestone for us. What we learned
We learned that building great software requires more than just code; it demands assertive, responsible, and coherent communication to keep the project moving forward. Beyond mastering new interpersonal dynamics, we acquired highly valuable technical skills in full-stack development, AI integration, and advanced Git workflows that will significantly strengthen our resumes. What's next for AppealForge
Our immediate next step is to refactor and improve the code's clarity to make it fully scalable and open-source friendly. We want to empower developers worldwide to contribute to this project, adapt the clinical RAG pipeline to their local medical guidelines, and deploy AppealForge to help patients in their respective regions.
Built With
- chromadb
- docker
- fastapi
- featherless.aiapi
- markdown
- oxlint
- pydinatic
- pymupdf
- python
- react
- react-markdown
- remark-gfm
- reportlab
- tailwind
- typescript
- uvicoin
- vite
Log in or sign up for Devpost to join the conversation.