Module 11: Industry-Level Projects — Collaboration, Documentation, Agile Workflows & Handoff | College AI Track

Learning Objectives

By the end of this module, you will be able to:

Plan multi-week AI/software projects with milestones, owners, dependencies, and definition of done.
Produce README, architecture notes, runbooks, CONTRIBUTING, and ADRs that onboard a stranger in <1 hour.
Execute Git: short branches, PRs, review norms, release tags.
Run lightweight agile: backlog, sprint goal, standups, review, retro—skip theater.
Hand off secrets hygiene, reproducibility, eval artifacts (Module 09), and team AI disclosure.

Deep Concept Explanation

11.1 Shipping mindset

Industry optimizes sustainable delivery and onboarding; courses often optimize solo rubrics. Converge: repos a hiring manager can clone, run, understand.

11.2 Vertical slices

One user path end-to-end (“question → retrieval → cited answer → logged latency”) before polishing secondary features—reduces integration risk.

11.3 RACI

Who is accountable for eval harness, infra, docs? Reduces duplicated or dropped work.

11.4 Definition of Done (AI projects)

Clean clone per README; tests/smoke pass; logs redact secrets; eval updated if prompt/corpus changed; docs updated; non-author PR review when team ≥2.

11.5 Documentation layers

README (quickstart, disclosure, troubleshooting); ARCHITECTURE (diagram, data flow, failure modes); CONTRIBUTING (branch, commits, review checklist); runbook (restart, keys, incidents); ADRs (decision, alternatives, consequences).

11.6 Agile kit (1–2 week sprints)

Backlog with acceptance criteria; one sprint goal; daily async/sync standup; demo software; retro with one committed improvement.

11.7 PR review dimensions

Correctness; readability; security; tests; docs; AI-generated hunks human-verified.

11.8 Branching

main demoable; short feature/*; tag v0.3-demo for judges.

11.9 Secrets and reproducibility

.env.example not .env; pin deps; optional containers; never commit keys.

11.10 Team AI charter

Allowed tools per artifact type; review rule for AI patches; README + appendix disclosure—no asymmetric secrecy.

11.11 Risk register

Likelihood × impact × owner × mitigation—technical, data, team, external (IRB delay).

11.12 Incident lite

Fallback if API down; postmortem: timeline, cause, preventive action—blameless.

11.13 Eval habits

eval/golden.yaml; optional CI smoke; changelog when prompts/index change.

11.14 Inclusion

Rotate scribe, presenter, reviewer; respect time zones and caregiving.

Code and Computational Examples

Key Example: Smoke tests fail fast on clean installs—pair with README commands your teammate can copy.

python

[object Object], pathlib ,[object Object], Path
,[object Object], importlib.util

,[object Object], ,[object Object],() -> ,[object Object],:
    spec = importlib.util.find_spec(,[object Object],)
    ,[object Object], spec ,[object Object], ,[object Object], ,[object Object],

,[object Object], ,[object Object],() -> ,[object Object],:
    root = Path(__file__).resolve().parents[,[object Object],]
    ,[object Object], (root / ,[object Object], / ,[object Object],).is_dir()

,[object Object], __name__ == ,[object Object],:
    test_import_package()
    test_fixtures_dir_exists()
    ,[object Object],(,[object Object],)

Describe without code: README skeleton sections; ADR template; user story Given/When/Then; YAML sprint fragment; PR checklist JSON; Mermaid in ARCHITECTURE.md; structured logging one-liner config; illustrative GitHub Actions YAML (push → checkout → setup-python → pytest).

Try This! Clone your repo into a new folder and follow README literally—every stumble becomes a doc fix.

Concept: main should always be demoable; experiments live on branches.

Docs are part of the product when failures are opaque.
Small PRs + review culture transfer to internships.
Eval and logging are scholarly reproducibility habits.

Key Takeaway

Treat README + ARCHITECTURE + eval/ as deliverables, not extras.
Pin environments and redact logs—AI systems fail in non-obvious ways.
PR hygiene (scope, review, CI smoke) signals professional maturity.
Risk registers and runbooks turn demo week panic into calm fallback paths.
Team AI disclosure norms prevent one silent draft from compromising everyone.

References and Cross-Modules

GitHub Docs; Accelerate (Forsgren et al.); Google Technical Writing; ADR patterns (search “architecture decision record”). Modules 01, 09, 10, 12.

Module 11: Industry-Level Projects — Collaboration, Documentation, Agile Workflows & Handoff