Course Concepts Map
The project covers all 10 course notebooks (Days 1a through 5b).
| Day | Notebook | Implementation | Wiki page |
|---|---|---|---|
| 1a | Foundational models | llm.py — tiered model registry with retry/backoff |
Model Registry |
| 1b | Multi-agent systems | orchestration.py — SequentialAgent pipelines, LoopAgent |
Agent Architecture |
| 2a | Agent tools & MCP | tools.py, mcp_server/ — clinical tools, FastMCP server |
MCP and A2A |
| 2b | Agent-as-tool & HITL | orchestration.py, human_in_the_loop.py — LongRunningFunctionTool |
Human-in-the-Loop Approval |
| 3a | Memory & state | memory.py — session/memory factories, state prefixes |
Memory Layers |
| 3b | Context engineering | context.py — token budgeting, compaction, boundary injection |
Memory Layers |
| 4a | Observability | observability.py, plugins.py — OpenTelemetry, Cloud Trace, structured logging |
Observability |
| 4b | Evaluation | eval/ — ADK EvalSet, tool trajectory + response match scoring |
Testing and Eval |
| 5a | Agent2Agent (A2A) | a2a_server.py — ASGI A2A server with agent card |
MCP and A2A |
| 5b | Deployment | deployment/ — Cloud Run, Vertex AI Agent Engine, GKE |
Deployment |
Evaluation rubric alignment
| Category | Points | Implementation |
|---|---|---|
| Technical Implementation | 50 | Multi-agent pipelines, MCP, 3-layer security, 4-layer memory, Pydantic, observability |
| Documentation | 20 | README, inline docstrings, architecture docs, this wiki |
| Core Concept & Value | 10 | Clinical intelligence with inspectable execution, tool, and evidence traces |
| Video Demo | 10 | End-to-end workflow demonstration |
| Writeup | 10 | Problem-solution-architecture-journey articulation |
Note: Capstone requirement Must demonstrate at least 3 of: ADK multi-agent, MCP server, Antigravity, security, deployability, Agents CLI. Clinical AI Kit demonstrates multi-agent (Agent Architecture), MCP (MCP and A2A), security (Security Layers), and deployability (Deployment).