Role-based multi-agent workflow for enterprise RAG use cases. Agents coordinate triage, research, drafting, critique, clarification, and validation using grounded tools.
| Agent | Responsibility |
|---|---|
| Triage Agent | Classifies query type, urgency, and required evidence |
| Research Agent | Retrieves policy and case context from hybrid search |
| Drafting Agent | Produces structured first response with citations |
| Critic Agent | Checks missing citations, ambiguity, and contradiction risk |
| Clarifier Agent | Requests user clarification when evidence is incomplete |
| Validation Tool | Calls backend APIs for policy lookup and data checks |
stateDiagram-v2
[*] --> Triage
Triage --> Research
Research --> Draft
Draft --> Critique
Critique --> Clarify: insufficient evidence
Critique --> Validate: grounded enough
Validate --> Final
Clarify --> Research
Final --> [*]
- LangGraph state management
- Tool schemas and parameter validation prompts
- Retryable backend tool-calling
- Policy lookup and data validation
- Reasoning-path and tool-trace summaries
- Safe refusal behavior when evidence is missing
Python · LangGraph · LangChain · Azure OpenAI · FastAPI · Pydantic · OpenTelemetry