Building dataset pipelines under free compute
What source-ID deduplication, supervised tokens, and packed windows say about preparation.
Commerce search and aggregation for Uzbek and regional product discovery.
ArzoMarket combines consumer aggregation with B2B discovery and analytics. It was originally developed as Findza and later renamed.
I founded ArzoMarket and lead it as CEO, and I built most of its technical side myself. The team has four members: me (founder and CEO), a CTO, a Pitch and Manager, and a Designer.
The same product can appear with inconsistent names and metadata across sources. Aggregation is useful when attribution, freshness, extraction, and matching are reliable.
Public commerce sources (planned)
Playwright indexing (direction)
Parsing · entity matching
FastAPI / MySQL
React / TypeScript
Initial product architecture and planned indexing direction; live source coverage is not implied.
The initial implementation uses a monorepo, FastAPI, React, TypeScript, MySQL, DDD-oriented backend structure, JWT, and an /api/v1 namespace. Earlier deployments used Vercel and Render. Later database experiments are history, not a replacement canonical stack.
Backend/API structure, auth integration, deployment, and indexing direction form the supplied work. Integration issues included auth responses, route handling, CORS, and frontend base-URL configuration.
The preferred direction is Playwright for publicly available dynamic product pages, with parsers, source attribution, respectful rates, and change detection. Candidate sources do not establish live integrations.
An earlier 98%+ entity-matching target is a goal, not measured accuracy. Evaluation needs labeled product pairs and precision/recall before a headline result is meaningful.
Experimental. Uzum, Olcha, Asaxiy, Wildberries, OLX, Telegram, Instagram, and wholesale channels are discussed sources; live coverage is not claimed. Affiliate/lead revenue, B2B subscriptions, promoted listings, and seller services are plans, not stated current revenue.
The crawler works, but some target sites block it with robots.txt rules and CAPTCHAs. We do not bypass either, so the project is pivoting to a different approach to getting product data.
A verified current demo, repository, screenshots, and coverage metrics have not been supplied.
An AI-first operating system on the Linux kernel: a Rust desktop shell and an agent daemon behind a permission broker and audit log, built with AI tools.
A language written in C11: typed compiler, optimizer, bytecode VM, tensors with autodiff, and a package manager. 10 versions and a full test suite.
Backend work for structured SAT practice and exam-like workflows.