Slaydar

Your closet as governed data in DataHub. An AI roasts your wear habits using only verified stats. No cap.

A wardrobe agent that models your clothes as governed data assets in DataHub — tracking real wear history, ownership, and staleness — so an AI can roast your habits with only verified numbers, and a buyer can trust a garment's condition on resale.

Built for Build with DataHub: The Agent Hackathon by Ishani Kathuria & Yasaa Usman Khan. 📄 Full submission writeup: DEVPOST.md · 🎥 Demo video: [PASTE DEMO VIDEO URL]

---

The idea in 30 seconds

Your closet has two hidden problems, and both are really metadata & provenance problems — DataHub's home turf:

1. You don't know your own wardrobe. You wear ~20% of your clothes 80% of the time, buy duplicates, and forget things exist.
2. Resale runs on blind trust. "Worn twice, like new" is unverifiable.

Slaydar models each garment as a DataHub Dataset entity with real ownership, usage stats, staleness flags, and cross-owner lineage. On top of the graph sit three AI agents (vision extraction, check-in matching, and a stat-backed roast persona). The roast is architecturally forbidden from citing any number that isn't in the DataHub record — a small, legible proof of an LLM grounded in governed metadata.

We don't just read DataHub — every catalog, check-in, staleness sweep, and resale writes back to the graph.

---

What it does

Step	What happens	DataHub write
Catalog	Upload closet photos or paste a product link → vision agent extracts category/color/material/brand/tags	creates a Dataset entity + Ownership + GlossaryTerms + properties
Check in	Photograph today's outfit → match agent finds the garment → you confirm	wear_count, last_worn_date, cost_per_wear, condition_score updated
Roast	Slaydar delivers one stat-backed line using only real stats	(reads the graph)
Flag	4+ wears in 7 days = overworn; 30 days untouched = never-worn	native Deprecation aspect
Resell	List a flagged item with a verified Condition Score; transfer ownership	new dataset URN + UpstreamLineage → live lineage graph

---

Architecture

Next.js app (frontend + runtime agents)      FastAPI service (the only DataHub writer)
 - upload / check-in / roast / listing UI --> - POST /garments               (create entity)
 - Gemini agents run in-app at runtime:       - POST /garments/{id}/checkin
     · vision extraction                      - POST /garments/evaluate-staleness
     · check-in matching                      - POST /garments/{id}/transfer-owner
     · roast generation                       - GET  /garments/{id}/lineage
                                              - GET  /closet/{owner_id}
                                                        |
                                              acryl-datahub Python SDK
                                                        |  (MetadataChangeProposals)
                                              DataHub GMS  (docker quickstart, v1.7.0)
                                              :9002 UI · :8080 GMS

Two services, one shared DataHub instance, a small REST contract (docs/api-contract.md) between them. Only /api talks to DataHub; the frontend talks only to /api.

How a garment becomes a real DataHub entity

No forked entity types — every garment is a DataHub Dataset (urn:li:dataset:(urn:li:dataPlatform:slaydar,<garment_id>,PROD)), and each concept maps to a built-in aspect it genuinely fits:

Garment concept	DataHub aspect
Owner (reassigned on resale)	Ownership
Style tags	GlossaryTerms
category / color / material / brand / wear_count / last_worn_date / cost_per_wear / condition_score / cataloged_date / status	DatasetProperties.customProperties
Overworn / never-worn	Deprecation
History across owners	new URN per ownership period + UpstreamLineage

Full detail in docs/datahub-schema.md.

---

Quickstart

Prerequisites: Docker, Python 3.10–3.12 (not 3.13+), Node 18+, a Gemini API key (free tier), and the DataHub CLI (pip install acryl-datahub).

1. Backend + DataHub

cd api
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# One command brings up the whole stack: DataHub -> glossary -> API -> tunnel -> seed
./scripts/demo_up.sh --seed

This starts DataHub (datahub docker quickstart, ~1 min), seeds the glossary + a coherent demo closet, and runs the API on :8000. DataHub UI is at http://localhost:9002 (datahub/datahub).

Running the API by hand instead: uvicorn app.main:app --reload --port 8000. If DataHub is unreachable the client runs in dry-run mode (logs the MetadataChangeProposal it would emit) so the API still boots for frontend work.

2. Frontend

cd web
npm install
# .env.local:
#   GEMINI_API_KEY=your_key_here
#   API_BASE_URL=http://localhost:8000
npm run dev

Open http://localhost:3000.

Run the tests

cd api && source .venv/bin/activate
pip install -r requirements-dev.txt
pytest            # hermetic: forced dry-run, never touches the shared DataHub

---

Demo flow

Mirrors the recorded video (see PLAN.md §7):

1. Upload closet photos → extraction agent tags items live.
2. Daily check-in → Slaydar roasts a repeat outfit, citing a real wear count.
3. Overworn / never-worn flag triggers a resale suggestion.
4. Paste a product link → resolves to a canonical item.
5. Close on the listing's Condition Score → then pull up the live DataHub lineage graph showing the garment's ownership chain. This is the moment that proves the DataHub integration is real.

---

Repo layout

Path	What	Owner
web/	Next.js 16 (TypeScript) frontend + runtime Gemini agents	Yasaa
api/	FastAPI service wrapping the acryl-datahub SDK	Ishani
docs/api-contract.md	REST contract between /web and /api	shared
docs/datahub-schema.md	DataHub entity/aspect mapping	shared
PLAN.md	Full build plan, schema rationale, locked agent prompt, demo script	shared
DEVPOST.md	Devpost submission writeup	shared

---

Tech stack

Frontend: Next.js 16 (App Router, React 19) · TypeScript · Tailwind v4 · Google Gemini 2.5 Flash (@google/genai) Backend: FastAPI · Python · acryl-datahub SDK · Pydantic · pytest Infra: DataHub v1.7.0 (datahub docker quickstart) · Docker · ngrok

---

Team

• Ishani Kathuria — DataHub modeling & FastAPI backend (/api)
• Yasaa Usman Khan — frontend & runtime agents (/web)