Strelec Labs
RUStart a project
Menu

AI-BUILT SOFTWARE RESCUE

Production Rescue

You built it with AI. I make it safe to run.

Claude, Cursor, Lovable and other AI tools can get a product surprisingly far. I take over when the codebase becomes hard to trust: audit the real state, fix the critical paths, launch safely and keep it alive.

  • AI-Built Codebases
  • Production Rescue
  • Auth / Payments / Webhooks
  • Deployment & Reliability
  • Ongoing Care

Remote · RU / EN

OWN SYSTEMS / ENGINEERING CASES

What rescue engineering looks like

Real work on my own systems: inherited state, retries, browser automation, AI execution and production verification.

01 / Backend · State machines

Retry-safe backend workflows

Regression tested; external operations mocked

Problem
Retries could repeat work or overwrite an order’s active state.
Engineering
Stable identities, durable state and guarded lifecycle transitions.
Result
Regression checks preserve the original order and prevent duplicate work.
Evidence
Regression suite · Lifecycle assertions

02 / Automation · Telegram

Deterministic browser automation

Live canary and replay verified

Problem
Browser delivery evidence and duplicate detection were unreliable.
Engineering
Exact conversation matching. A durable queue. Confirmed outgoing text and a cleared composer.
Result
A live self-canary sent once. Replaying it created no second message.
Evidence
Browser canary · Duplicate replay · Queue state

03 / AI workflows · Execution gates

AI tied to real execution

Synthetic flows tested; runtime deployed

Problem
An assistant could claim work had started without creating an executable task.
Engineering
Deterministic intent checks against the existing order state, with execution evidence and QA gates.
Result
Controlled tests submit no task when blocked, and submit once with review when executable.
Evidence
Synthetic regressions · QA assertions · Deployment verification

SENIOR ENGINEERING APPROACH

Understand first. Preserve what works. Fix what matters.

No blind rewrite. I establish the actual system state, protect working paths, fix the highest-risk failures and verify the result in the environment that matters.

I take over AI-built and inherited projects

I read the code AI produced, map the real dependencies and identify what is canonical before changing the system.

I trace business-critical paths end to end

Auth, payments, webhooks, database state, runtime, Linux, network and deployment are one system. The symptom is only the starting point.

I leave the system maintainable

A focused fix, regression checks, deployment evidence, rollback and a clear handoff. After launch I can keep the system healthy month to month.

ENGINEERING METHOD

  1. 01

    INSPECT

    Read the code. Inspect logs, runtime and network.

  2. 02

    REPRODUCE

    Trigger the original failure. Capture evidence.

  3. 03

    ROOT CAUSE

    Trace the failure to its actual cause.

  4. 04

    FIX

    Apply the smallest safe change.

  5. 05

    VERIFY

    Retest the symptom, regression and deployment.

  6. 06

    DELIVER

    Return the checked result with evidence and rollback.

A fix is complete only after the original failure has been checked again.

PROBLEMS I SOLVE

Your app already exists. Now it has to survive reality.

Perfect for products assembled with Claude, Cursor, Lovable, Replit or mixed human/AI development — especially when real users, payments and production expose the hard parts.

01

AI-built app audit

You have a working demo or early product, but no longer trust the code. I review architecture, auth, secrets, permissions, payments, webhooks, data flows and deployment risk — then rank what must be fixed before growth.

Claude · Cursor · Lovable · Existing code

Start a project
02

Rescue sprint

Fix the critical paths that are blocking launch or revenue: auth, roles, API contracts, Stripe/payment flows, webhook idempotency, queues, database state and recurring production bugs.

Root cause · Targeted fixes · Regression checks

Start a project
03

Production launch

Turn a codebase that works locally into a controlled production system: CI/CD, Docker/VPS/Vercel/Railway, environment separation, secrets, backups, monitoring, health checks and rollback.

Deploy · Observability · Backups · Recovery

Start a project
04

Keep it alive

Ongoing technical care after launch: monitor failures, review risky AI-generated changes, maintain dependencies and infrastructure, handle incidents and keep recovery paths tested.

Monitoring · Maintenance · Incident response

Start a project
05

Existing system takeover

A previous developer left, several branches disagree, or nobody knows what is really deployed. I reconstruct the actual state, preserve valid work and create a safe path forward.

Codebase mapping · Release truth · Handoff

Start a project

ENGINEERING STACK

Tools chosen for the system and the problem.

  • TypeScript
  • JavaScript
  • Node.js
  • Bun
  • Python
  • Next.js
  • React
  • Linux
  • Docker
  • GitHub
  • Railway
  • Vercel
  • API / Webhooks
  • Telegram
  • AI / LLM
  • Automation
  • systemd
  • Networking

STARTING POINTS

Start with diagnosis. Continue only if it makes sense.

These are starting points, not fixed quotes. Rescue work starts with a bounded diagnosis so you know what is wrong before paying for a larger rebuild or sprint.

Codebase Triage

from $99

Focused diagnosis of one blocking production problem.

Production Audit

from $350

Risk-ranked review of an AI-built or inherited application.

Rescue Sprint

from $900

Critical fixes, regression checks and controlled production delivery.

Production Care

from $299/month

Monitoring, maintenance and incident support after launch.

START A PROJECT

What did AI build — and what stopped being simple?

Describe what the product does, how it was built, what you no longer trust and what must work in production. A repository or public URL is enough to start the conversation.

  1. Send the brief
  2. Receive your request reference
  3. Continue the same request in Telegram

No account needed. Do not send credentials. I start with evidence and a bounded scope, not a blind rewrite.

Add contact, links or an error excerpt

Share a task, not credentials. Leave passwords, API keys and private data out of the brief.