Skip to content
All projects

Case study · Product

ThreadFlank

A fashion app where shoppers see clothes on themselves before they buy.

Indian fashion brands put their catalog on it, shoppers try the clothes on using AI, and when they like something they go to the brand's own website to buy it.

Timeline
Jul 2026 – present
My role
Full stack: AI pipeline, backend and both web apps.
ThreadFlank home page
  • 2

    apps: brand panel + shopper app

  • 69

    features shipped

  • 37

    built and tested

  • 9

    on the roadmap

  • Next.js
  • NestJS (Fastify)
  • TypeScript
  • PostgreSQL
  • Prisma
  • Redis
  • BullMQ
  • Socket.IO
  • Gemini
  • OpenRouter
  • Replicate
  • LangChain
  • Pinecone
  • Razorpay
  • Jest

Screenshots

Inside ThreadFlank

  • ThreadFlank screenshot 1
  • ThreadFlank screenshot 2
  • ThreadFlank screenshot 3
  • ThreadFlank screenshot 4
  • ThreadFlank screenshot 5
  • ThreadFlank screenshot 6

How it works

From brand catalog to buy tap

  1. 1

    A brand joins

    Uploads its clothes (one by one, or hundreds at once from a spreadsheet) and sets up its store page.

  2. 2

    A shopper finds the brand

    Browses brands and collections, follows the ones they like, joins the waitlist for new launches.

  3. 3

    The shopper tries clothes on

    The app shows the garment on the shopper's own photo using AI, so there is less guessing about fit and look.

  4. 4

    The shopper buys

    Tapping Buy takes them to the brand's own website. ThreadFlank counts the tap so the brand sees what works.

  5. 5

    The brand learns and grows

    The brand sees views, try-ons, saves and buy taps, and uses AI tools to write its marketing copy.

The product

Two apps, one product

Every feature, grouped by app. Working = built and tested; Built = exists, not re-tested recently.

For brand owners and their team. Where a brand manages its clothes, launches, numbers, AI tools and payments. Brands are the supply side: no brands and clothes, no reason for shoppers to stay.

Team and security

Who can use the brand's account, and what each person can do.

  • Sign up and approvalBuilt

    A brand confirms its email, then a ThreadFlank admin approves it before shoppers can see it.

  • Login and password resetBuilt

    Sign-in, sign-out, forgot-password and change-password.

  • Invite teammatesWorking

    Teammates get an email, choose their own password and join the brand's account.

  • Roles: Owner, Manager, ViewerWorking

    Managers do everything except buy credits and manage the team; Viewers can only look.

  • Store settingsBuilt

    Name, story, logo, banner, accent colour, announcement, website and physical-store flag.

  • GST number and badgeWorking

    Format-checked GST number shows a 'GST registered' badge to shoppers.

Catalog

Built so one brand can manage 500+ pieces and the platform can hold 1000+ brands.

  • Add, edit and delete piecesWorking

    Title, price, original price, colours, sizes, SKU, category, occasions, collection, stock, live/draft.

  • Up to 8 photos per pieceBuilt

    Uploads are shrunk automatically, or brands paste links to photos they host.

  • Search, filters and sortingWorking

    Filter by category, collection, status, featured, stock or 'not ready for try-on'.

  • Smooth long listWorking

    Loads a screenful at a time, so it stays fast with 500 pieces.

  • Bulk actionsWorking

    Publish, hide, delete or set collection/category for up to 500 pieces at once.

  • CSV importWorking

    Checks every row, imports the good ones, reports the bad ones; Shopify-style columns and a template.

  • CSV exportWorking

    Download the whole catalog as a spreadsheet.

  • CollectionsWorking

    Group pieces under a name; shoppers see them as cards and filters.

  • Featured pieces and live/draftWorking

    Show favourites first and keep unfinished pieces hidden.

Dashboard and insights

Real shopper activity, so brands can see whether ThreadFlank works for them.

  • Dashboard overviewWorking

    Views, try-ons, saves, buy taps, conversion and followers for 7/30/90 days, compared with the period before.

  • Top piecesWorking

    Ranks pieces by try-ons, then saves and shares, then buy taps.

  • Insights pageWorking

    Day-by-day chart, a view → try-on → save → buy funnel, activity by category, top 25 pieces.

  • Download the numbersWorking

    Export the top-pieces table as a spreadsheet.

  • Activity trackingWorking

    Repeat views of the same page within 30 minutes count once.

  • Try-ons from real rendersBuilt

    Counted only after the AI finished and the shopper paid.

Drops and launches

Build excitement for a new collection.

  • Schedule a dropWorking

    Title, collection, description, launch date and banner.

  • Countdown and statusWorking

    Days to go, and scheduled / live / finished.

  • WaitlistWorking

    Shoppers join; the brand sees how many are waiting.

  • Drop linked to a collectionWorking

    Shoppers get a 'Shop this collection' button on launch day.

  • Launch notificationsWorking

    Waitlist and followers get one in-app notification when the drop goes live.

AI Studio

AI tools that save a brand's marketing team time. Credits pay for them.

  • AI writer: 7 kinds of copyWorking

    Descriptions, Instagram captions, announcements, emails, SEO, WhatsApp broadcasts, short titles: three options each.

  • Tone, language and lengthBuilt

    Luxe, playful, minimal, festive or bold; English, Hindi or Hinglish.

  • Uses the brand's real detailsBuilt

    The AI gets the piece's real title, price, colours, sizes and occasions.

  • Fair chargingBuilt

    2 credits, taken only after the text is produced.

  • Saved copy libraryWorking

    Save, browse by type, copy again or delete (up to 500 items).

Billing and credits

How the platform earns money from brands.

  • Credit packsWorking

    Starter (50 credits, Rs 499), Growth (200, Rs 1,499), Scale (600, Rs 3,999), GST included.

  • Pay online with RazorpayWorking

    UPI, cards and net banking; only the owner can pay.

  • Credits are never lostWorking

    Background callback adds credits if the browser closes; wrong amounts add nothing; never twice.

  • Promo codesWorking

    Percentage off, bonus credits or both, with use limits and expiry.

  • Purchase historyWorking

    Date, pack, credits, amount and payment status.

Getting seen

Other ways a brand reaches shoppers.

  • Runway biddingBuilt

    Brands bid for a featured slot on the Discover page.

  • Try-on widget for the brand's siteBuilt

    A snippet that adds try-on to the brand's own website, limited to its listed domains.

  • Marketing and pricing pagesBuilt

    Public pages that explain ThreadFlank to brands.

Engineering

The hard parts

  • AI try-on on a digital twin

    • Multi-person try-on on a reusable digital twin of the shopper (measurements + selfie).
    • Cost-aware routing across Gemini, OpenRouter and Replicate with automatic fallback.
  • Catalog built for 500+ pieces per brand

    • CSV import checks every row and understands Shopify-style columns; 500 rows imported in under a second in testing.
    • Bulk publish, hide, delete or re-tag up to 500 pieces; the list loads a screenful at a time.
  • Payments that never lose credits

    • Razorpay's background callback still adds credits if the brand closes the browser; a wrong amount adds nothing, and credits are never added twice.
    • Tested with test keys and signed test callbacks.
  • Numbers brands can trust

    • Store views, piece views, try-ons, saves, shares and buy taps are tracked; repeat views within 30 minutes count once.
    • A try-on counts only after the AI finished and the shopper paid for it.
  • AI Studio for marketing copy

    • 7 kinds of copy in English, Hindi or Hinglish, written from the piece's real title, price, colours and sizes.
    • Costs 2 credits, taken only after the text is produced; failures are free.
  • Real-time social backend

    • Socket.IO, Redis Pub/Sub, BullMQ, rate limiting and JWT power DMs, group outfit planning, voice/video calls and the Discover feed.
    • Launch day is automated: waitlist and followers get one in-app alert when a drop goes live.

Architecture

How it fits together

Clients

Shopper app

Brand panel

Try-on widget on brand sites

API

NestJS (Fastify) · TypeScript

Data and real time

PostgreSQL + PrismaRedis (Pub/Sub, cache)BullMQ jobsSocket.IO real timePinecone (RAG)

External services

Gemini · OpenRouter · ReplicateLangChainRazorpay

Roadmap

Not built yet

  • Email and phone-push alerts (today, notifications are in-app only)
  • Reviews and questions on pieces
  • Shopify and WooCommerce sync (spreadsheet import for now)
  • Monthly subscription plans and GST tax invoices for brands
  • A screen for staff to create promo codes
  • Motion video of a piece (screen exists, video service not connected)
  • Creators and influencers features
  • Shopper coupons and offers
  • Bot and abuse protection (Cloudflare planned)

Honest notes

  • Payments were tested with signed test callbacks, not a real payment.
  • A real try-on using the paid AI image services was not run end to end in tests.