rauhan.sheikh
← Back to Work Directory
PROFESSIONAL[MBZ]2026

MarketBaaz

Helps Amazon sellers decide what to sell: a Chrome extension and web app that turn marketplace data into clear insight, and never show a number the evidence can't support.

Visit Live System
4 data types
Data Engine
Products, keywords, brands and sellers, gathered by the extension itself
24 country sites
Reach
Amazon marketplaces worldwide, each with its own currency
Every ~2 days
Shipping Pace
A new extension version roughly every two days for five weeks
Sole engineer
Ownership
Extension, web app, backend and infrastructure

Engineered With

Next.jsReactTypeScriptNode.jsMongoDBRedisBullMQPythonChrome ExtensionsRazorpayFirebase Auth

The Problem & Context

Sellers choose products on guesswork or pay for tools whose numbers they can't interrogate. A research tool that invents confident-looking figures is worse than none: someone acts on them and loses real money. The challenge was to be useful and honest at the same time.

The Messy Part (Behind the Build)

Real-world data is hostile. Marketplaces differ in layout and currency, pages change under you, listings vary by size and colour, and the source site pushes back on aggressive reading. Getting one trustworthy dataset out of that, at a speed that doesn't get anyone blocked, is most of the work.

Architecture & Design Decisions

A Chrome extension reads the pages a user already visits and talks to a backend API; nothing proprietary ships to the client. The backend (Next.js, TypeScript, Node.js) stores everything in MongoDB as a single source of truth, runs background jobs on Redis and BullMQ, and serves estimates that are computed ahead of time by a trained model rather than on every request. Payments, authentication, email and an admin console complete the platform, and the whole stack is self-hosted on our own VPS.

What it is

MarketBaaz is an Amazon seller-intelligence platform, formerly BrandForge. It has two halves that share one backend:

  • A Chrome extension that adds research directly onto Amazon pages while a seller browses.
  • A web app with product, brand and seller research, keyword tools, an AI brand-building suite, a marketplace, billing and a full admin console.

It works across 24 Amazon country sites with correct local currencies, and I built and run it as its only engineer. The web app was a waitlist when I joined; I took it public, built the extension end to end, and then shipped a new extension version roughly every two days for five weeks.

An estimator that knows what it doesn't know

The most important design decision was a refusal:

If the evidence isn't there, the system says so.

Instead of a fixed formula, estimates come from a trained machine-learning pipeline that is retrained on a schedule, versioned, and checked against its previous output. When a product doesn't have a reliable signal, users see an explicit state such as "Not enough data", never a fabricated figure. A ranking layer on top is configurable and carries a confidence value, so a result backed by plenty of evidence is distinguishable from a thin one.

Fast at scale

The extension gathers the data itself as people browse, and MarketBaaz has grown into a data engine covering products, keywords, brands and sellers. Keeping it quick as that data grew meant fixing real bottlenecks: a database sort that ran out of memory and crashed product listing was moved into the database itself, targeted indexes and caching were added, writes became bulk and repeat-safe, and page weight was cut (the main logo went from 243 KB to about 6 KB).

Built to be operated

  • Rollouts: features ship dark and are enabled per plan or audience from the admin console.
  • Billing: subscription changes, renewals, retries, tax invoices and discount codes.
  • Support: in-product problem reporting that creates tickets in the admin support centre.
  • Migration: I moved the live system to new infrastructure and a new domain, with staged user emails, a rollback-ready release and the old domain kept working during the transition.

What I'd Do Differently Today

I would separate workloads onto their own hosts from day one. Running application, workers and databases together made every incident bigger than it needed to be, and the migration that fixed it was avoidable.

Related Work & Systems

PERSONAL

BusinessOS

Commercial-grade financial operating system for SMEs with ledger-first accounting, immutable transaction journals, integer minor-unit money calculations, and multi-tenant audit logs.

PROFESSIONAL

Microsoft Teams & Calendar Engine

Hiding deep Microsoft Graph API complexity behind an automated calendar & meeting sync engine for 20K+ EdTech users.