Capstone Starter Template Guidance

You already have working building blocks from Weeks 1–4. Assemble them into a single evaluator rather than rewriting from scratch.

Reuse map

Capstone need	Course artefact
Document-aware chunks + metadata	labs/01-chunking
Hybrid retrieval / RRF	labs/01-chunking/src/retrieval.py
Storage abstraction	labs/02-storage (PgVectorStore, graph)
Modular skills + router	labs/03-skills
Golden dataset + CI gate	labs/04-evaluation

Suggested student project layout

dfd-compliance-evaluator/
├── src/
│   ├── skills/          # from Week 3 (or thin wrappers)
│   ├── router/
│   ├── retrieval/
│   ├── storage/
│   └── evaluation/
├── data/
│   ├── policies/
│   ├── fixtures/
│   └── golden_dataset.jsonl
├── tests/
├── scripts/ or terraform/
├── .github/workflows/eval.yml
└── README.md

Minimum acceptance path

1. Ingest policy docs → enriched corpus (Week 1).
2. Index with chosen store (Week 2) and justify the choice.
3. Route DFD input through modular skills (Week 3).
4. Emit structured ScoreComplianceOutput-style JSON.
5. Pass labs/04-evaluation deterministic gates (or an extended golden set) at the Capstone thresholds (>90% precision / groundedness on your suite).

Tips

• Keep syntax and graph skills deterministic.
• Treat DFD JSON as untrusted input.
• Version thresholds in config; do not hard-code magic numbers in CI YAML only.