How hastimal/gemini-healthcare-agentic-platform works
This is an agentic healthcare research and discovery platform. A user question is planned into searches, handled by research agents and MCP-backed healthcare sources, then ranked and turned into an evidence-grounded answer with citations and limitations. The graph also preserves the documented model-portability and evaluation capabilities. Most agent and grounding implementations were not sampled, so workflow links below rely on the README’s explicit architecture; unsampled internals are represented at their exact directory paths.
Main entry points: healthcare_workflow.py Healthcare workflow · healthcare.py Healthcare MCP client · agent.py Healthcare Research Agent · model_factory.py Gemini and Gemma
Components 24
User Experience
- Research demo
ui/streamlit_app.py
Agent Workflow
- Search Planner Agent
agents/search_planner/agent.py - Query fan-out
search/query_fanout.py - Healthcare Research Agent
agents/healthcare_research/agent.py - Healthcare workflow
agents/workflows/healthcare_workflow.py
Search and Evidence
- Healthcare MCP client
mcp_services/clients/healthcare.py - Search MCP tools
mcp_services/search_server/tools.py - Research MCP tools
mcp_services/research_server/tools.py - FHIR MCP tools
mcp_services/fhir_server/tools.py - NPPES provider search
connectors/provider_search/client.py - PubMed search
connectors/pubmed/client.py - FHIR R4 connector
connectors/fhir/client.py - Evidence retrieval
search/retrieval.py
Grounded Answering
- Evidence ranking
grounding/ranking.py - Evidence & Answer Agent
agents/evidence_answer/agent.py - Grounded synthesis
grounding/answering.py - Citations
grounding/citations.py
Models and Evaluation
- Gemini and Gemma
llm/model_factory.py - Deterministic evaluation
evaluation/orchestrator.py - Evaluation metrics
evaluation/metrics - Evaluation reports
evaluation/reporting.py
Other components
- Healthcare user
- Cited recommendations
- Healthcare data services
Connections 27
- Healthcare user → Research demo: asks question
- Research demo → Healthcare workflow: starts workflow
- Healthcare workflow → Search Planner Agent: runs first
- Search Planner Agent → Query fan-out: builds search plan
search/query_fanout.py - Healthcare workflow → Healthcare Research Agent: runs research
- Healthcare Research Agent → Healthcare MCP client: uses MCP retrieval
- Healthcare MCP client → Search MCP tools: dispatches search
- Healthcare MCP client → Research MCP tools: dispatches research
- Healthcare MCP client → FHIR MCP tools: dispatches FHIR
- Search MCP tools → NPPES provider search: retrieves providers
- Research MCP tools → PubMed search: retrieves literature
- FHIR MCP tools → FHIR R4 connector: delegates searches
- NPPES provider search → Healthcare data services: queries NPPES
connectors/provider_search/client.py - PubMed search → Healthcare data services: queries PubMed
connectors/pubmed/client.py - FHIR R4 connector → Healthcare data services: queries FHIR endpoint
connectors/fhir/client.py - Healthcare Research Agent → Evidence retrieval: retrieves evidence
- Healthcare workflow → Evidence ranking: ranks evidence
- Healthcare workflow → Evidence & Answer Agent: invokes answer agent
- Evidence & Answer Agent → Grounded synthesis: grounds response
- Grounded synthesis → Citations: adds citations
- Grounded synthesis → Cited recommendations: returns recommendations
- Gemini and Gemma → Search Planner Agent: powers agent
- Gemini and Gemma → Healthcare Research Agent: powers agent
- Gemini and Gemma → Evidence & Answer Agent: powers agent
- Healthcare workflow → Deterministic evaluation: evaluated by
- Deterministic evaluation → Evaluation metrics: computes checks
- Deterministic evaluation → Evaluation reports: produces results