This project demonstrates a deterministic multi-agent workflow using LangGraph.
Two agents collaborate:
- Research Agent: Collects and stores external knowledge using web search
- Summary Agent: Uses Retrieval-Augmented Generation (RAG) to produce grounded executive summaries V
User Query -> Research Agent -> Vector Store -> Retrieval -> Summary Agent
- LangGraph-based orchestration
- Real web search integration
- Proper RAG implementation
- Memory via vector database
- Tool calling via LLM-bound tools
- Guardrails for unsafe queries and uncited summaries
- File-system logging for traceability
- Executive-grade summaries
- Create virtual environment
- Install dependencies
- Add API keys in
.env - Run
python main.py
Example query:
Latest trend in RAG
Example output (shape, abbreviated):
EXECUTIVE SUMMARY:
The field of Retrieval-Augmented Generation (RAG) saw significant advancements in January 2025 with the publication of 10 key research papers. These papers cover a range of topics such as integrating graph-structured data into RAG systems, developing lightweight RAG systems for resource-constrained environments, and enhancing security measures against adversarial attacks. Additionally, new frameworks like Agentic RAG and TrustRAG aim to improve the retrieval process and protect RAG systems from potential threats. The research also delves into best practices for designing RAG systems, evaluating their performance, and enhancing retrieval and reasoning capabilities. Overall, these papers provide valuable insights into the latest developments in RAG technology.
Sources:
1. https://example.com/source-one
2. https://example.com/source-two
LangGraph enables explicit control over agent execution order, shared state, and safety boundaries -- critical for production agent systems.