Agenda — Day 5 Part 1: End-to-End Capstone Project¶
The capstone is where two weeks of concepts become a single working system. You will design, build, evaluate, and present an LLM application that integrates at least four of the major components covered in this course.
Session goals¶
By the end of this session you will have:
- A deployed, working LLM application connected to real APIs
- A documented architecture with rationale for every design decision
- A quantitative evaluation suite with at least two metrics
- A 5-minute demo ready to present to the group
Schedule¶
| Time | Activity |
|---|---|
| 0:00–0:20 | Project brief + scope selection |
| 0:20–0:50 | Architecture design session (whiteboard / doc) |
| 0:50–2:30 | Implementation sprint |
| 2:30–3:00 | Evaluation + metrics run |
| 3:00–3:30 | Demo prep + presentation |
| 3:30–4:00 | Group demos + feedback |
Required components (choose at least 4)¶
- [ ] RAG pipeline (chunking + embedding + retrieval)
- [ ] LLM API calls with proper error handling and retries
- [ ] Streaming response delivery
- [ ] Function calling / tool use
- [ ] LangGraph agent with conditional routing
- [ ] Async FastAPI service
- [ ] Caching layer (exact-match or semantic)
- [ ] Evaluation suite (RAGAS, custom metrics, or LangSmith evals)
- [ ] Cost tracking and budget guard
- [ ] Deployment (Modal, Fly.io, or Docker + uvicorn)
Deliverables¶
- Working code — runnable from a
README.mdwith two commands - Architecture diagram — hand-drawn, Mermaid, or any tool
- Eval results — numbers, not vibes
- 5-minute demo — live or recorded