Unit 5 — Multi-Agent Systems
Overview
Design and orchestration of multi-agent systems for scientific research, from LangChain to Google ADK 1.0 with A2A protocol.
Topics
Frameworks Covered
| Framework | Focus |
|---|---|
| LangChain | Chains, tools, LCEL |
| LangGraph | Stateful graphs, cycles |
| CrewAI | Role-based collaboration |
| Google ADK 1.0 | Production-ready agents |
| SmolaAgents | Lightweight agents (HF) |
Advanced Topics
- A2A Protocol (Agent-to-Agent) — Cross-framework communication standard
- RAG & GraphRAG — Retrieval-Augmented Generation with graphs
- Mem0 + ChromaDB — Persistent agent memory
- Production: FastAPI, OpenTelemetry, model routing
Key Notebooks
| Notebook | Topic |
|---|---|
| U5_01 | LangChain fundamentals |
| U5_02 | LangGraph stateful agents |
| U5_03 | CrewAI multi-role teams |
| U5_04 | Google ADK 1.0 + A2A + Gemma4 |
| U5_05 | RAG and GraphRAG |
| U5_06 | Persistent memory (Mem0) |
| U5_07 | Production: FastAPI + observability |
| U5_08 | Multi-provider model routing |
| U5_09 | SmolaAgents (HuggingFace) |
U5_04 Highlights — Google ADK + A2A
The most advanced notebook in the unit covers:
LlmAgent,SequentialAgent,ParallelAgent,LoopAgentBuiltInCodeExecutor— Code execution in secure sandboxBaseToolwith strict Pydantic typing- Session checkpoints and recovery
- AgentCard and A2A discovery protocol
- Cross-framework client with Bearer authentication
- Local fallbacks for exhausted API quotas