Tana Bana

Project Dashboard
Logo
Login Page
Architecture

##Inspiration Traditional artisans create some of the world’s most beautiful crafts, yet remain invisible online. What inspired us was a simple truth: the internet demands typing, editing, and digital skills — but artisans speak through their hands, not keyboards. We wanted to build an interface that respects their reality. Not an app they must learn, but an app that learns them. Tana Bana is the attempt to weave human stories, culture, and AI into one continuous thread.

What it does?

Tana Bana lets an artisan speak in their own language and automatically generates: Clean product descriptions Pricing guidance SEO-ready tags And most importantly — a shoppable video reel The buyer sees a vertical reel feed (like Instagram Reels), taps a product, and checks out instantly. No typing, no editing, no digital literacy needed. Speak → AI → Shoppable Reel in 60 seconds.

How we built it?

We built a complete voice-to-commerce pipeline: Chirp STT v2 Converts noisy, multilingual artisan speech into accurate text. Gemini Flash 2.5 Generates titles, descriptions, stories, hashtags, SEO keywords, and pricing suggestions. Veo 3 Creates short, cinematic product reels from a single image. Next.js + PWA For instant mobile access on low-end phones with optimistic UI. Node.js + GCP Cloud Run Backend APIs for products, users, checkout, and media processing. Google Cloud Storage To store reels, images, and serve fast streaming. The entire system is optimized for unstable networks and budget Android devices — the real devices artisans use.

Challenges we ran into?

1) Voice Noise & Accent Variation Bazaar noise and strong local accents confused transcription. Solution: Prompt tuning + domain vocabulary protection + language detection.

2) Video Generation Memory Crashes Veo reels were heavy and crashed workers. Solution: Switched to async processing + GCS streaming to avoid memory overflow.

3) Multilingual Data Clean-Up Hindi–Hinglish–Bhojpuri mixed inputs broke prompts. Solution: Input normalization + forced structured output templates.

4) Poor Network Conditions Actions felt frozen on 3G. Solution: Optimistic UI + background syncing.

Accomplishments that we're proud of Achieved full end-to-end flow: voice→description→reel→listing→checkout Built an interface that artisans can use with zero typing Generated professional product reels from a single photo Made a functional PWA usable on low-network rural regions Created a new category: Voice-to-Reel Commerce

What we learned AI is powerful, but only if the human input is respected. Real users (artisans) don’t want more buttons — they want less thinking. Multilingual models require careful guardrails to avoid hallucinations. Decoupling heavy AI tasks from user flow is the difference between a demo and a usable product.

##What's next for Tana Bana? Launch pilots with artisans in Jaipur, Varanasi, Kutch, and Puri Recommendation engine for reels Voice-based pricing negotiation between buyer and seller Weekly auto-generated marketing reels for top products Cultural provenance badges powered by AI verification Scale to Indonesia, Mexico, Morocco, Philippines — global artisan clusters Build the world’s largest culture commerce graph.

About the Project Tana Bana is built on a simple belief: Craft deserves a digital home as beautiful as the hands that create it. We created a system that lets artisans enter the global market using the only tool they’ve always had — their voice.

Craft → Story → AI → Commerce. A new economy woven from old roots.