Live on Google Play & webVideo-commerce marketplace

Local shops thatsell through video.

A hyperlocal video-commerce marketplace: neighbourhood stores sell through shoppable reels, nearby shoppers buy from the video, and riders deliver — with AI that writes listings and makes the reels.

Product: Vidco vidco.in↗
Client: Hyperlocal commerce · India
Role: Full-stack engineer
Surfaces: Merchant app · Rider app · Shopper web · AI studio

The brief

Small shops don't have marketing teams, but they do have phones. The idea: let every local store sell the way creators do — with short shoppable videos — and let customers nearby discover and order from them.

That turned into a three-sided marketplace — shoppers, merchants and riders — on one platform, with a control centre that can reconfigure the whole product without a release.

Merchant app

Onboarding with KYC, catalogue, Reel Studio, AI create hub, orders, payouts and analytics.

Rider app

Go online, accept timed delivery offers, run trips with proof of delivery, track earnings.

Shopper website

Store-first discovery, reels, product pages, cart and checkout.

Control centre

Operations, finance, moderation and CMS — plus live control of theme, languages, flags and integrations.

System map

How it fitstogether.

Every surface talks to one platform core. This is the shape of the system as it runs in production.

The hard parts

What made ithard — and how.

The problems that don't show up in a screenshot, but decide whether a product survives real users.

Shoppable reels, rendered on the server

The problem

Merchants design overlays — prices, tags, animated callouts — on a phone. The final video must look exactly like the preview, on every device.

The approach

One overlay description drives both the live preview in the app and a server render: each overlay becomes an image, timed and animated onto the video, then transcoded to adaptive streaming in a dedicated worker.

Video pipelineffmpegHLS

Streaming without easy downloading

The problem

Merchant videos are their marketing. They should play anywhere but not be trivially scraped.

The approach

Stream manifests are rewritten on request so every segment points to its own short-lived signed URL.

Signed URLsPrivate streaming

Dispatching the nearest rider

The problem

Each order needs one rider, fast — but riders decline, go offline or ignore offers.

The approach

Each task is offered to the single nearest online, verified rider, preferring riders whose zone covers the store. Declines and timeouts cascade to the next rider, and a one-minute sweep re-offers anything stuck.

Geo queriesDispatch engineIdempotent sweeps

Stock that never drifts

The problem

Carts reserve stock; abandoned checkouts must give it back without double counting.

The approach

Checkout places a timed hold on stock, and a sweep releases expired holds before the database expires the reservation record — so inventory always balances.

Inventory holdsTTL

One subscription model, three stores

The problem

Merchants subscribe through Apple, Google Play or Razorpay depending on device.

The approach

Every provider's events are mapped onto one shared set of billing events, so plans and AI credits behave identically.

Billing abstractionIAP

AI in production

An AI studio for shops without a studio

A provider-pluggable AI module turns a merchant's phone photos into a ready-to-sell listing and a shoppable reel. Jobs run in the background, credits are reserved up front and refunded on failure, and plan limits and cost tracking keep it sustainable.