MediMemo – Conversational Medical Records
Built with Pramod Thebe at the Multimodal Frontier Hackathon, where it won Best Use of the Railtracks Agentic Framework. Healthcare data is fragmented and hard to interpret. MediMemo lets a clinician or patient upload lab reports, prescriptions, and notes, then ask questions in plain language and get answers with citations back to the source document. Under the hood is a hybrid RAG pipeline: 500–800 token chunks with overlap, BM25 plus vector search, and a reranker, which together improved retrieval relevance by roughly 30% over vector search alone.
Project Features
Multimodal Document Intake
PDFs, scanned reports, and notes are classified, chunked, and indexed by Railtracks agents so every later answer can point back to a page.
Hybrid Retrieval
BM25 keyword search and vector search run together, then a reranker orders the candidates. Roughly 30% better relevance than vectors alone in our tests.
Cited Answers
Every response is grounded in retrieved chunks and shows its citations, which matters more in medicine than anywhere else.
Production Deployment
Containerized and deployed to Azure Container Apps with Terraform, with persistent patient history across sessions.
Project Info
03/2026
Backend & Retrieval Pipeline
Team of 2
Winner · Best Use of Railtracks
Technologies Used
- • Next.js, assistant-ui
- • FastAPI, Server-Sent Events
- • Railtracks ADK agents
- • Hybrid RAG: BM25 + vector + rerank, 500–800 token chunks
- • Weaviate / Qdrant, MinIO
- • Docker, Azure Container Apps, Terraform
Project Gallery
Interested in This Project?
I'm always excited to discuss my projects and share insights about the development process. Feel free to reach out if you'd like to know more!