Projects
The projects are the runnable heart of the course. Each one changes one harness lever, asks you to inspect the trace, and leaves behind a policy or artifact you can carry forward.
Project Path
| Project | Focus |
|---|---|
| P01: Agent Trace | Read the agent loop from its event trace |
| P02: Model Routing | Right-size the model to the task |
| P03: Retrieval | Decide when search tools earn their slot |
| P04: Decomposition | Split large work into smaller runs |
| P05: Memory | Add durable context to cut re-discovery |
| P06: Safety | Bound the agent's blast radius and sandbox it |
| P07: Critic Capstone | Add evaluation and compose the full harness |
| P08: Dynamic Workflows | Move orchestration into reusable workflows |
| P09: Model Routing Benchmark | Select the most appropriate model |
| P10: Indexing Agent History | Add a database/index over conversation traces |
| P11: Subagents | When do subagents make sense? |
How To Work
- Open the project page.
- Open the
starter/files in GitHub. - Predict the trace.
- Run the smallest experiment.
- Compare with
solution/. - Save the result table or policy.
The full project READMEs live in the repo: projects/.