Skip to Content
RAG

RAG

Rosetta uses semantic search to retrieve relevant information from your notes, uploaded documents, and external medical databases.

Architecture

The system converts text into high-dimensional vectors that capture meaning rather than just matching keywords. When you ask a question, Rosetta finds the most semantically similar content from your available sources to generate an answer.

Semantic vs Keyword

Keyword Search (Traditional):

Query: "heart failure treatment"
Matches: Exact text "heart failure" AND "treatment"
Misses: "HFrEF management", "cardiac dysfunction therapy"

Semantic Search (Rosetta):

Query: "heart failure treatment"
Matches: Any semantically similar concepts
  ✓ "HFrEF management"
  ✓ "cardiac dysfunction therapy"
  ✓ "GDMT for reduced EF"

Source Types

Rosetta organizes knowledge into two scopes:

  1. Local: Sources specific to the current note relative to the patient context.
  2. Account: Guidelines, protocols, and papers you have saved to your personal library, available across all notes.

PubMed Integration

The system directly integrates with the PubMed database to retrieve evidence-based literature.

  • Search: Queries are converted to search terms for the PubMed database.
  • Retrieval: Abstracts and metadata are fetched for the top results.
  • Ingestion: Retrieval results are processed into the RAG system for the agent to use in citations.

Usage Examples

Evidence-Based Treatment (HFrEF)

  1. Physician adds a “GDMT” PDF to their Account library.
  2. In a note, they ask: “heart failure reduced ejection fraction treatment”.
  3. The system retrieves guidance on ACE-I/ARB, beta-blockers, and SGLT2i usage.
  4. The agent generates a plan with citations from the uploaded PDF.

Literature Review (AFib)

  1. The agent searches PubMed for “atrial fibrillation anticoagulation 2024”.
  2. It retrieves recent abstracts.
  3. The agent generates a summary with the latest scoring criteria.

Institutional Protocols

  1. User uploads a hospital sepsis protocol to their Account library.
  2. Any note can now query: “UCSF sepsis bundle timing”.
  3. The agent returns the specific timing requirements defined in your protocol.

Next: Workflows | API

Last updated on
EditorWorkflows