Aydin Aliyev
Bu sayt Azərbaycan dilində də var.Azərbaycanca oxu
Этот сайт есть и на русском.Читать по-русски

How I can help

Three ways to work together. Each says what you get, how it runs, and which work proves it.

Open to one audit or build at a time

  1. Audit an AI-built app

    For founders whose product was written mostly by AI and is about to take real users or real money.

    What you get

    • Recon of the codebase and how it's deployed
    • Findings I have reproduced, ranked by what they could cost you
    • A security review and an architecture review
    • A fix plan with an estimate for each fix

    How it runs

    AccessReconFindingsReportWalkthrough

    Proof: two audits
  2. Build it with an agent team

    For a product that has to exist soon, built with agentic engineering: coding agents work from specs, and one senior engineer is accountable for every line.

    What you get

    • Specs with acceptance criteria, written with you
    • Working software in production, not a prototype
    • The repository with its tests, decision records and task history

    How it runs

    SpecAgentsGateShip

    Proof: RentPilot
  3. Join your team, on contract

    For teams that need a senior full-stack engineer who reviews a lot of AI-written code.

    What you get

    • Features shipped end to end, from schema to screen
    • Daily review of your team's pull requests, human and AI
    • Remote from Baku (GMT+4), used to working with teams in the US
    • React
    • Next.js
    • Node
    • Go
    • GraphQL
    • Postgres
    • AWS
    Proof: ten years of work

What a finding looks like

One entry from an audit report, with the client's details blacked out. The real report has one of these for every issue I could reproduce.

F-03HighFixed, retested

Model output reaches users without escaping

Where
The summary endpoint in redacted
What
A summary written by the model is rendered as HTML. Text planted in a user's own record can come back as a working link in someone else's dashboard.
Reproduced
Yes, on staging, with a record created for the test. redacted
Fix
Render summaries as text, allow a short list of Markdown marks, and strip links the model didn't get from the source record.
Estimate
Half a day, plus a regression test

Sample based on a real finding class; wording, file and severity are illustrative.

How it starts

Send me a short note: what you're building, where it is now, and what worries you about it. I'll write back with questions, and if it's a good fit, a one-page scope with a price and dates. Nothing starts until you've said yes to it.

Work with me