Submission Checklist¶

Use this checklist to verify your capstone project before submission. Every item marked required must be complete. Items marked stretch are optional but improve your portfolio.

Code quality¶

[ ] Required — No hardcoded API keys. All secrets loaded via os.getenv().
[ ] Required — requirements.txt or pyproject.toml with pinned versions.
[ ] Required — README.md with: what the project does, setup instructions (2–3 commands), example usage.
[ ] Required — At least one working FastAPI endpoint with Pydantic request/response models.
[ ] Required — Error handling for API failures (try/except, HTTP error codes, meaningful messages).
[ ] Stretch — Dockerfile or deployment config (Modal, Fly.io) included.
[ ] Stretch — .env.example showing required environment variables (values blanked out).

Architecture and integration¶

[ ] Required — At least 4 course components integrated (see 00-agenda component list).
[ ] Required — Architecture diagram included (DESIGN.md or README.md).
[ ] Required — Async patterns used correctly (AsyncOpenAI, await, not blocking the event loop).
[ ] Stretch — LangSmith tracing enabled and at least one trace screenshot in the README.
[ ] Stretch — Cost tracking: total tokens and estimated cost logged per request.

Evaluation¶

[ ] Required — At least one quantitative metric defined and measured.
[ ] Required — Eval script that runs against the live API and prints results.
[ ] Required — At least 10 test inputs (20+ for a credible eval).
[ ] Stretch — RAGAS evaluation (faithfulness + answer relevancy) for RAG projects.
[ ] Stretch — Comparison of two approaches (e.g., chunking strategy A vs B, or with/without reranker).

Presentation¶

[ ] Required — 5-minute live demo or recording covering: problem, architecture, demo, eval results.
[ ] Required — Can explain every design decision made.
[ ] Required — Can identify the biggest current limitation and one concrete improvement.
[ ] Stretch — Latency breakdown: time to first token, total time, tokens/second.

Repository structure¶

Your submission repository should look like this:

my-capstone/
├── README.md              ← Setup, usage, results summary
├── DESIGN.md              ← Architecture decisions and data flow
├── requirements.txt       ← Pinned dependencies
├── .env.example           ← Required env vars (values blanked)
├── app.py                 ← Main FastAPI application
├── ingest.py              ← Data ingestion script (if RAG)
├── eval.py                ← Evaluation script
└── docs/                  ← Source documents for RAG (if applicable)

Self-assessment rubric¶

Rate yourself honestly. This is for your own learning, not grading.

Area	1 — Not done	2 — Partial	3 — Complete	4 — Polished
Integration depth	1–2 components	3 components	4 components	5+ components
Code quality	Runs with errors	Runs on happy path	Handles errors	Production-ready
Evaluation	No metrics	One metric, few examples	Two metrics, 10+ examples	RAGAS or ground truth
Architecture clarity	No diagram	Rough sketch	Clear diagram + rationale	Matches code exactly
Demo quality	Doesn't run	Runs once	Reliable, shows edge cases	Recorded + narrated

The most important thing

A working system that does one thing well is more valuable than a half-built system that claims to do five things. Ship the core pipeline first, then add features.

05-presentation-guide