How kkm121/multi-agent-research-assistant works
This is a research-assistant application: a user submits a topic, optionally with a document, through the Chainlit chat; the UI streams progress and a final report from the FastAPI service. The LangGraph workflow generates expert personas, gathers facts, outlines and synthesizes a report, then critiques it and may revise it. The README also documents multi-user chat persistence and quantitative Ragas evaluation. The graph’s workflow and service links are shown in source; optional provider and database internals are represented only where excerpts support them.
Main entry points: agents.py Agent Operations · app.py Chainlit Chat · graph.py LangGraph Workflow · main.py FastAPI Service
Components 17
User Experience
- Chainlit Chat
frontend/app.py - Chat Persistence
frontend/app.py
API Service
- FastAPI Service
backend/main.py
Research Workflow
- LangGraph Workflow
ai_core/graph.py - Persona Generation
ai_core/graph.py - Fact Research
ai_core/graph.py - Report Outline
ai_core/graph.py - Report Synthesis
ai_core/graph.py - Critique Revision
ai_core/graph.py - Agent Operations
ai_core/agents.py - Research Models
ai_core/models.py - Search Tools
ai_core/tools.py
Evaluation
- Ragas Evaluation
ai_core/ragas_eval.py
Other components
- Researcher
- Semantic Scholar
- DuckDuckGo Search
- LLM Providers
Connections 23
- Researcher → Chainlit Chat: submits research
frontend/app.py - Chainlit Chat → Chat Persistence: reads and writes
frontend/app.py - Chainlit Chat → FastAPI Service: streams request
frontend/app.py - FastAPI Service → Chainlit Chat: mounts UI
backend/main.py - FastAPI Service → LangGraph Workflow: invokes workflow
backend/main.py - LangGraph Workflow → Persona Generation: starts at
ai_core/graph.py - Persona Generation → Fact Research: then researches
ai_core/graph.py - Fact Research → Report Outline: then outlines
ai_core/graph.py - Report Outline → Report Synthesis: then synthesizes
ai_core/graph.py - Report Synthesis → Critique Revision: then critiques
ai_core/graph.py - Critique Revision → Report Synthesis: revises if needed
ai_core/graph.py - Persona Generation → Agent Operations: generates personas
ai_core/graph.py - Fact Research → Agent Operations: gathers facts
ai_core/graph.py - Report Outline → Agent Operations: generates outline
ai_core/graph.py - Report Synthesis → Agent Operations: drafts report
ai_core/graph.py - Critique Revision → Agent Operations: reviews report
ai_core/graph.py - Agent Operations → Research Models: uses typed models
ai_core/agents.py - LangGraph Workflow → Research Models: uses workflow state
ai_core/graph.py - Agent Operations → Search Tools: runs queries
ai_core/agents.py - Search Tools → Semantic Scholar: searches papers
ai_core/tools.py - Search Tools → DuckDuckGo Search: falls back to
ai_core/tools.py - Agent Operations → LLM Providers: calls model providers
ai_core/agents.py - Ragas Evaluation → LangGraph Workflow: evaluates output
ai_core/ragas_eval.py