AI / ML system
CatalogWala
Product Builder · Solo · Vernacular AI for trades
AI-powered portfolio/catalogue for Indian blue-collar tradespeople — clusters gallery photos into projects, writes vernacular titles/descriptions, and exports WhatsApp cards + PDFs.
Status: Working · vernacular catalogue AI
Train
Inference
Output
Model pipeline
Research → result
Lab-style panels for training, inference, and outputs.
System design
Overview
A catalogue product for painters, plumbers, carpenters and other trades — turn messy phone galleries into a clean, shareable work portfolio in Hindi, Gujarati, and English.
What I built
- Multilingual onboarding (Hindi, Gujarati, English)
- Smart photo import: gallery, date-range scan, in-app camera
- On-device pipeline: quality scoring, MobileNet embeddings, clustering
- AI enrichment (Cloud Functions → Anthropic Claude) for titles/descriptions
- Shareable WhatsApp cards, PDF catalogue, public portfolio link
- WorkManager pipeline: Import → Analyze → Cluster → Enrich → Sync
- Pro subscriptions via Play Billing (monthly/yearly)
What's unique
- Built for low-literacy / vernacular-first users (large touch targets, warm Material 3 UI)
- On-device ML: blur/exposure scoring, embeddings, clustering, duplicate detection
- Cloud enrichment via Claude for titles/descriptions in 3 languages + WhatsApp/PDF outputs
Why it matters
- Solves a real Indian MSME gap — tradespeople rarely have polished portfolios
- Photos stay on-device until publish/enrich — privacy-aware by design
- End-to-end product path: import → analyze → cluster → AI enrich → share → Pro billing
Craft
How the intelligence was built
AndroidJavaXMLMVVMHiltRoomWorkManagerOpenCVTFLiteFirebaseClaudePlay Billing
Open the research