Healthcare RAG Platform — Overview

Complexity tier: Intermediate Business Problem: (fill in — e.g., clinical staff / patients need fast, accurate answers from large volumes of unstructured healthcare documents — policies, care guidelines, FAQs — without waiting on a human expert.)

Why this project

Leverages existing healthcare domain background (DaVita Kidney Care experience) to produce a credible, differentiated interview narrative — not a generic tutorial clone.

Scope (v1)

• Ingest a small set of sample clinical/policy documents
• Chunk → embed → store in a vector DB
• Retrieve relevant chunks for a query
• Generate grounded answers via Gemini on Vertex AI

GCP Services Used

• Vertex AI (embeddings + Gemini generation)
• (Vector store — decide: Vertex AI Vector Search / AlloyDB / other — log decision in decisions.md)
• Cloud Run (serving, once past prototype stage)
• Cloud Storage (source documents)

AI Components

• Embedding model
• Retrieval strategy (top-k, reranking — TBD)
• Generation / grounding via Gemini
• (Later) Evaluation harness for answer quality

Stretch Goals

• Add an ADK agent layer on top (e.g., a triage agent that decides which knowledge base to query)
• Multi-document-type support (structured + unstructured)
• Add basic evaluation pipeline (faithfulness, relevance)
• Add observability/tracing

Interview Relevance

Directly demonstrates: RAG system design, Vertex AI hands-on experience, healthcare domain credibility, and (once agentic layer is added) agentic system design — all top-priority FDE interview topics.