MediVise

Medical documents are difficult for patients to parse; the product needed structured workflows for documents, medications, and appointments plus an AI layer that could summarize and answer questions without crossing into clinical advice.

Make medical document information more accessible while keeping AI outputs bounded, source-grounded, and clearly non-diagnostic within a capstone scope.

• Educational capstone scope—not a clinical or FDA-regulated product
• Safety boundaries required for any AI-generated health-related content
• Latency-sensitive RAG chat over user-uploaded documents of varying quality and format

• Built a React web app with per-user auth, Supabase/Postgres document metadata, and CRUD workflows for appointments, medications, and wellness features.
• Engineered a FastAPI backend with phi-4-mini RAG over ingested medical documents, tuning hybrid (semantic + keyword) retrieval for lower latency and more relevant answers.
• Cached third-party lookups (RxNorm medication data, location geocoding) and framed all AI output as assistive—clearly non-diagnostic.

• Prioritized document summarization, RAG chat, and core workflows over peripheral wellbeing features
• Used hybrid retrieval instead of semantic-only search to improve recall on medical terminology
• Framed all AI outputs as assistive summaries and Q&A—not diagnostic or prescriptive guidance

• Hybrid retrieval tuning for relevance and latency on document Q&A
• Cached RxNorm and geocoding calls to reduce redundant third-party requests
• Safety-bounded AI framing with explicit non-diagnostic disclaimers in product copy

Built a capstone prototype combining document accessibility workflows with a grounded RAG assistant—demonstrating full-stack delivery, retrieval tuning, and responsible AI boundaries in a healthcare-adjacent domain.

• Not shipped; educational capstone prototype only
• Not intended for diagnosis, treatment decisions, or medical advice
• RAG evaluation and failure-mode testing remain limited to manual review

• Expand automated evaluation for summarization and RAG answer quality
• Harden auth boundaries and document access policies before any broader release

• React
• FastAPI
• PostgreSQL (Supabase)
• phi-4-mini RAG
• Hybrid retrieval
• RxNorm API