Clinical guideline Q&A with LlamaIndex

The scenario

CareGuide shows the full LlamaIndex lifecycle. Guidelines are parsed and enriched with literature abstracts, chunked and embedded into a vector index, and queried through a query engine that answers only from retrieved evidence. Evaluations measure faithfulness, an API and a simple UI expose it, and input/output guardrails sit at the API boundary. It is an educational tool, not medical advice — and it says so.

System architecture

High-level flow of CareGuide, from input to output.

What you will learn

• Document parsing and node construction in LlamaIndex
• Vector indexes and query engines
• Faithfulness evaluation for high-stakes domains
• Guardrails at the API boundary

What you will build

• Guideline + literature ingestion
• Grounded query engine with sources
• FastAPI service and Streamlit UI
• Dockerised CI/CD deployment

Build plan

1. Parse and enrich the corpus
2. Build the index and query engine
3. Evaluate faithfulness
4. Add guardrails and the UI
5. Containerise and deploy

Tech stack