Skip to content

Wearly AI References and Course Context

Companion document to Wearly_Product_Brief.md. Organizes the course context, instructor expectations, and research/reference strategy for Wearly AI.

1. SEIS 666 Project Context

Wearly AI is being developed for:

  • SEIS 666 — Digital Transformation with AI
  • Final Project — Build Something That Reasons
  • Track — Track B: Agentic AI System

The project must demonstrate:

  • a working system,
  • a multi-step workflow,
  • at least two tools or data sources,
  • a documented decision process,
  • error handling,
  • a before/after comparison,
  • GitHub documentation, and
  • a 10-minute demo.

The final deliverable should be a working product prototype — not only a PDF, slide deck, or concept explanation. The app is the product; documents support it.

2. Official Course and Project References

3. Course Principles Relevant to Wearly AI

The following course principles directly shape Wearly AI's design philosophy:

  • Build, do not only theorize. A working artifact is the deliverable.
  • Context is the product. Quality of recommendations comes from organized context — calendar, weather, wardrobe, fit, history — not from clever prompts alone.
  • Structure beats volume. Well-structured small data and rules outperform large, unstructured input.
  • AI systems should reason with organized context, not random prompting. The reasoning trace must be visible and defensible.
  • The strongest projects show how AI changes the workflow before and after. A clear delta between manual and agent-assisted styling is required.
  • The final app remains the main product. Documentation, books, and graphs support the app; they do not replace it.

4. Intelligent Book References

The Intelligent Book is not the product itself. It is a structured documentation and reasoning layer that supports the working app.

The Intelligent Book for Wearly AI should explain:

  • product vision,
  • user problem,
  • agent workflow,
  • styling knowledge base,
  • wardrobe intelligence,
  • fit / body profile logic,
  • calendar and weather context,
  • shopping gap logic,
  • before/after demo,
  • privacy and security, and
  • product roadmap.

References

5. Agent Skills and Meta-Skills References

The Skills folder should package reusable workflows, rules, decision logic, references, and validation checks for the Wearly Styling Agent.

The Wearly skill should eventually include:

  • SKILL.md,
  • occasion rules,
  • weather rules,
  • fit and silhouette rules,
  • wardrobe filtering rules,
  • color coordination rules,
  • wear history rules,
  • shopping gap rules, and
  • privacy guidelines.

References

6. Lessons from Class Demo Feedback

Synthesized from prior class project feedback:

  • The instructor prefers a working MVP, not just a PDF or slide deck.
  • The demo should show the system working live.
  • The project should clearly show why it is an agent.
  • The project should have a clear target audience.
  • The project should include a strong before/after comparison.
  • The project should explain the decision process, not only show the final output.
  • The project should address privacy and security when using personal data such as calendar, location, wardrobe photos, measurements, preferences, and shopping behavior.
  • The project should mention testing or UAT, even if full UAT is future work.
  • The strongest projects feel like real products or serious prototypes, not toy examples.

7. Research Strategy for Styling Logic

Wearly AI's recommendation logic should be evidence-informed, not based only on generic prompting.

It should combine:

  • fashion recommendation system research,
  • color harmony principles,
  • body-positive fit and silhouette rules,
  • occasion and formality rules,
  • weather and location constraints,
  • wardrobe availability,
  • wear history,
  • user rejection reasons, and
  • user feedback.

Styling is not an exact science. Wearly AI should combine evidence-informed rules, user preferences, context, and feedback — rather than claim there is one universally correct outfit.

8. Academic and Expert References

The canonical, structured references document is docs/evidence-and-references.md — see also Intelligent Book chapter book/12-evidence-and-references.md. Nine source categories, mapped to each Skill rule pack, with a verification-status table.

Source categories Wearly draws on

# Category Status
1 Fashion recommender systems General practice
2 Outfit compatibility General practice
3 Color harmony General practice (warm/cool framework, Itten / Munsell / CIE pending verification)
4 Body-shape-aware / fit-aware styling General practice (industry heuristic, not scientific taxonomy)
5 Explainable recommendation systems General practice (Miller 2019 pending verification)
6 Human-centered AI / personalization Well-known framework (Microsoft AI Guidelines, Google PAIR pending verification)
7 Wardrobe management / digital closets General practice
8 Personalization and user feedback General practice
9 Privacy and personal data Well-known framework (GDPR, privacy-by-design pending verification)

Honesty contract

  • No fabricated citations. Categories are named when they're textbook common knowledge; specific authors, papers, and findings appear only after verification.
  • Body-shape category labels are industry heuristic, not scientific taxonomy — Wearly never claims a user "is" a body shape; the label is a user-declared proportion preference.
  • Color palettes are design-informed conventions, not deterministic claims — the language is "tends to read well," not "is correct."
  • Body-positive language is enforced at runtime in code and statically across documentation.

14 pending verification items

Concrete searches a future contributor can complete — full list in docs/evidence-and-references.md §7. Highlights:

  1. Recommender Systems Handbook (Ricci, Rokach, Shapira) — edition.
  2. Polyvore-dataset paper or successor (outfit-compatibility benchmark).
  3. Itten color wheel — verified publisher / edition.
  4. Munsell color system — verified standard reference.
  5. CIE color spaces — CIELAB / CIELUV standard numbers.
  6. Tim Miller, "Explanation in Artificial Intelligence" — verified citation.
  7. Microsoft AI Guidelines for Human-AI Interaction (18 guidelines) — URL + version.
  8. Google PAIR Guidebook — URL + current version.
  9. GDPR — specific article numbers for user-rights claims in book/10-privacy-security.md.
  10. Ann Cavoukian privacy-by-design framework — title + year.
  11. Peer-reviewed study (if any) of skin-tone × garment-color preference.
  12. Peer-reviewed work on body-shape taxonomy validity (or counter-perspectives).
  13. Inclusive-fashion / fat-positive design research.
  14. A fashion-XAI user study (HCI / IUI venue).

When a citation is verified, the workflow is in docs/evidence-and-references.md §8.

9. How These References Should Be Used

Reference set Role in Wearly AI
SEIS 666 project references Define what the project must demonstrate (rubric, deliverables, tracks).
Course references Define the philosophy — reasoning, context, structure, and working systems.
Intelligent Book references Guide the documentation structure that supports the app.
Agent Skills references Guide the reusable skill package layout and content.
Styling research references Guide the recommendation logic so it is evidence-informed, not generic.

The final app must remain the main product. References, documents, books, and graphs support the product — they do not replace the working demo.

10. Immediate Next Step

After this file and Wearly_Product_Brief.md are created, the next step is to:

  1. Ask Claude Code to read both files,
  2. Review the current repository, and
  3. Create a phased master implementation plan — without immediately editing code.

The implementation plan should align with the roadmap in Wearly_Product_Brief.md (Phase 0 → Phase 6) and respect the principles, references, and instructor expectations summarized here.

Companion to Wearly_Product_Brief.md. Together these two documents form the foundation for the Wearly AI build.