End-of-year project (PFA) · team of 3
Mohammed VI University Hospital — Neonatology
Rebuilt the department’s application as microservices, with an AI assistant connected to patient records through an MCP server.
What I did
-
AI assistant: a Mastra agent (Gemini) connected to patient records through an MCP server (Spring AI); a RAG over pgvector (bge-m3 embeddings) lets it find the right record field before reading or searching the data.
-
Document scanner: a Python pipeline (LLM + RAG) that turns unstructured Word reports into structured patient records, with patients anonymised (spaCy NER) before anything reaches the LLM.
-
Containerised microservice architecture (Docker Compose, nginx): backend, doctor and admin frontends, AI assistant, audio transcription and report generation.
-
Integrated the audio transcription (Whisper) and PDF report generation (LaTeX) services into the backend and frontend.
-
CI/CD with GitHub Actions: a reusable pipeline that builds Docker images and deploys automatically to a VPS, greatly speeding up the team’s iterations.