ALMS

AI Image Enhancement Pipeline

Artisan uploads a phone photo. The image is sent to the Gemini Vision API with a structured prompt requesting background removal guidance and quality assessment. A server-side Sharp pipeline then applies background removal (rembg), auto-levels, and sharpening based on the AI's quality score. The enhanced image is uploaded to Cloudflare R2 alongside the original.

Multilingual Content Generation

The artisan speaks or types a short description in any supported language. Gemini translates and expands this into a structured product listing: title, short description, material callouts, and care instructions — in all 4 target languages simultaneously. The prompt is engineered to preserve cultural context and craft terminology.

PWA Offline Architecture

Draft listings are stored in IndexedDB via a service worker. When connectivity is restored, the sync worker uploads images to R2 and pushes listing data to the API. Conflict resolution uses a last-write-wins strategy with a device timestamp.

// Service worker: queue failed uploads
self.addEventListener('fetch', (event) => {
  if (event.request.method === 'POST' && isListingAPI(event.request.url)) {
    event.respondWith(
      fetch(event.request.clone()).catch(() => {
        return queueForSync(event.request);
      })
    );
  }
});

Deployment

Next.js on Vercel Edge CDN with ISR for marketplace pages. MongoDB Atlas for data. Cloudflare R2 for image storage with public CDN URLs.

CI/CD

GitHub Actions — lint + build on PR, Lighthouse CI for PWA score check, auto-deploy on merge to main.

Monitoring

Vercel Analytics. Lighthouse PWA score tracked on every deploy. MongoDB Atlas alerts for slow queries.

v2.0 — Q1 2027

• —WhatsApp integration for artisan onboarding and listing updates
• —AI-powered demand forecasting for seasonal crafts
• —Direct payment integration with UPI

Artisan listing dashboard — offline-capable PWA