Azamat Egamberdiev

I turn a business's scattered data into numbers it can trust.

Analyst and systems builder at a cosmetics distributor trading across the CIS: reconciliation, tax and e-invoicing, demand forecasting, warehouse control — and the tooling that keeps AI coding agents honest while they build them.

The code runs a live business, so it stays private. Each system has a public page: the problem, how it is built, and how its answers are verified.

Role
Analyst and systems builder
Markets
Uzbekistan and other CIS markets
Based in
Tashkent, Uzbekistan
Works in
Python, JavaScript, SQLite, 1C
Now studying
Machine learning
Open code
6 lab tools

Selected work

Operations · since Jun 2026, active

Marketplace Control Tower

Pulls marketplace, tax, bank and 1C data into one database, reconciles the sources against each other, forecasts demand and answers questions over read-only SQL.

“A wrong total looked exactly as convincing as a right one.”
Read the case studyPython · SQLite
flowchart LR
  S["Marketplaces · fiscal receipts<br/>e-invoices · bank · warehouse<br/>1C in two countries"] --> C["Collectors<br/>source health + totals check"]
  C --> D[("One database<br/>barcode spine")]
  D --> E["Reconciliation · demand<br/>procurement · cash"]
  E --> O["Owner reports<br/>+ daily alerts"]
  D --> B["Analyst bot<br/>read-only SQL"]

Lab

Small tools with readable code, on synthetic data. Clone them and run them.

How I work

My part is the part that decides whether the numbers are right: framing the problem, defining the invariants a correct answer has to keep, designing the checks, and auditing what comes back.

Coding agents (Claude Code and Codex) write most of the code to that specification, and several of these projects carry their own audit harnesses for exactly that reason.

Want a walkthrough of any of these?

I can show a system end to end, or give read access to a project.

eazamat360@gmail.com