Play 28
Knowledge Graph RAG
High✅ Ready
Graph-enhanced retrieval with entity extraction and relationship mapping.
Combines knowledge graphs with vector search for richer, more contextual answers. Azure Cosmos DB Gremlin API stores entities and relationships extracted from documents. When a query arrives, the system retrieves both vector-similar chunks AND graph-connected entities, producing answers that understand relationships between concepts. Especially powerful for complex domains (legal, medical, scientific) where understanding entity relationships is critical.
Architecture Pattern
Graph RAG: entity extraction + relationship mapping + vector fusion
Azure Services
Azure OpenAI (gpt-4o)Cosmos DB (Gremlin API)Azure AI SearchContainer Apps
DevKit (.github Agentic OS)
- agent.md — root orchestrator with builder→reviewer→tuner handoffs
- 3 agents — Graph RAG Builder (gpt-4o), Reviewer (gpt-4o-mini), Tuner (gpt-4o-mini)
- 3 skills — deploy (103 lines), evaluate (105 lines), tune (101 lines)
- 4 prompts — /deploy, /test, /review, /evaluate with agent routing
- .vscode/mcp.json — FrootAI MCP with Cosmos Gremlin + OpenAI inputs + envFile
TuneKit (AI Config)
- config/openai.json — entity extraction + answer gen prompts
- config/graph.json — entity types, relationship types, traversal depth
- config/guardrails.json — hallucination prevention via graph grounding
- evaluation/eval.py — Entity accuracy >85%, Relationship precision >80%
Tuning Parameters
Graph traversal depth (1→3 hops)Entity type definitionsRelationship weight scoringVector vs graph fusion ratioAnswer grounding styleGraph update frequency
Estimated Cost
Dev/Test
$150–300/mo
Production
$2K–6K/mo