A RAG assistant inside a patient records system that answers physicians’ questions with cited sources, and says so when it can’t answer.
RAG
Guardrails
LLM evaluation
React
Use case
Physicians need quick answers from dense patient and clinical records. In healthcare an answer is only useful if you can trace it back to its source, and the system has to know when not to answer.
Constraints
Patient data is sensitive, so inputs and outputs are validated and protected.
The assistant lives inside the existing records workflow, not in a separate tool.
A confident wrong answer is worse than no answer.
What I built
The retrieval pipeline: chunking, embeddings and semantic search over source content.
Grounded generation, with the sources cited next to every answer.
Guardrails: input and output validation, sensitive-data protection, grounded-answer checks and low-confidence fallbacks.
Evaluation and monitoring of answer relevance, faithfulness and accuracy in production.
The full-stack platform around it: REST API integrations, front-end modules and a shared React component library.
Trust comes from being able to check the answer, not from how confident it sounds.