Have you built an application with AI?

The prototype works. Are you ready to take it beyond a hobby project and start handling users, their data and payments? Or have you already launched, but performance or security problems are appearing and more prompts only make things more complicated?

I can help you work through this. I review the code, configuration and application behaviour. The report clearly explains what needs fixing, the specific actions to take and their order. I also propose a maintenance plan: how to verify future changes and what to monitor to reduce the risk of similar problems.

From my experienceA user bought 100 credits. They received 200.

A user bought 100 credits.The payment completed successfully.

  1. First confirmation+100
  2. Repeated confirmationAfter a timeout+100

The user received200 creditsinstead of the 100 purchased.

The application added credits each time it received confirmation of the same payment. A user who knew the relevant link could use it to add any number of credits.

Audit and report
from PLN 1,000 excl. VAT
We agree the price of each audit individually. We agree the scope when signing the contract. Fixing the problems found is quoted separately.
Receive your report within 5 business days.
After the contract is signed and the agreed access is provided. If the audit requires a broader scope of work than usual, we agree the delivery date individually.
We discuss the report in a 60-minute meeting.
I explain the vulnerabilities found, recommended fixes and how to avoid similar mistakes as the application develops.

Is this audit right for you?

Who is this review for?

You have a prototype and want to welcome your first users.

You built the application in Lovable or Bolt, or with help from Cursor, Claude Code or Codex. Perhaps your data is in Supabase. The screens look good, but you have mostly tested scenarios where everything succeeds. Before launch, you want to check edge cases, security, denied access and what happens if the database goes down.

You have launched and the first problems are appearing.

You already have users. The application slows down, a payment does not grant access or a fix breaks another feature. You want to get things under control, understand what causes these problems and how to resolve them.

You hired someone to build an application and wonder if you could have done it yourself with AI.

A developer, agency or software house is delivering your project. You want to know what you are getting beyond the finished screens. I check the agreed requirements, permissions and error handling. The report gives you specific findings to discuss with the contractor.

Who is it not for?

You are building only for yourself.

You are learning or exploring an idea. The application does not collect other people’s data or take payments, and you are not planning to release it yet. A full audit is usually unnecessary at this stage.

You want to find out whether people will use your application.

You have an idea, but do not yet know whether it solves a problem for future users or whether they would pay for it. Start by showing them a prototype and gathering feedback. An audit can assess how the application works, but it cannot establish demand for the product.

You want your website to stop looking like an AI template.

The initial excitement about your project has worn off. You see the same layout and components as on other AI-built websites. I also refine their design, adapting typography, composition and interface elements to your brand and checking them on desktop and mobile.

See the AI slop removal service

What does an application audit look like in practice?

You show me the application and explain what it should do and what concerns you. We agree the scope, then I check the selected scenarios in the running application, code and configuration.

  1. We agree the scope and prepare access.

    On a 30-minute call, we discuss the application, user roles and reported problems. When signing the contract, we agree the scope, exclusions and price. You provide the agreed access to the repository, configuration and test environment.

  2. I test the behaviour and compare it with the code.

    I use accounts with different permissions. I check what happens when a user interrupts a payment, tries to open someone else’s invoice or the application fails to save their changes. I review the code and configuration and measure performance.

  3. I document findings, evidence and priorities.

    For confirmed defects, I record the conditions, reproduction steps, impact and recommended fix. I distinguish these findings from hypotheses and state what could not be checked. The report also explains how to verify the fixes.

  4. We discuss the report and next steps.

    In a 60-minute meeting, I explain the findings and answer your questions. We discuss repair priorities, how to verify fixes and how to maintain the application. Repairs are quoted separately.

What do you gain from the audit?

The report explains what needs fixing, where to start and how to verify the results. You can work from it yourself or give it to a developer.

Findings backed by evidence

For each confirmed defect, I describe the impact and how to reproduce or measure it. I distinguish established causes from issues that need further investigation.

Specific recommendations and repair priorities

I explain what needs changing and why. You can see which problems put data or payments at risk and which improvements can be scheduled later.

A way to verify fixes and maintain the application

I describe how to check repairs, what to test after future changes and what to monitor while the application runs. This gives you a reference for further development.

60 minutes to discuss the report

We walk through the findings and recommendations. I explain what they mean for your application, answer your questions and help establish where to start.

What do clients come to me with?

New changes kept breaking features that had worked before.

A client was building an application with AI, but each new change broke working features. They had lost control of the work. With no earlier versions of the code or configuration, I could not establish what had changed since the last working version or restore it.

I helped organise subsequent iterations and record changes so they could be tracked and earlier versions restored. This was hands-on support with building the application, without an audit report.

Users paid, but the application did not grant access.

A client came to me because paying for a subscription did not give users access. Payments went through and subscriptions were recorded in the database, but the application did not recognise the users’ entitlement to use the service.

I found a problem in the endpoint configuration: payment data was being sent to the wrong environment. There were also problems with webhooks, the notifications from the payment provider. I identified where communication between the provider and the application needed correcting.

Changing an ID exposed other customers’ invoices.

A client showed me an application whose interface displayed only each user’s own data. I demonstrated that changing or guessing an identifier in a request to the server was enough to retrieve other customers’ data.

I also found unprotected PDF invoices. They were available online through links containing a simple ID, without checking permission to download them. Both the data served by the application and the documents themselves needed access protection.

An AI-built importer could not handle the product catalogue.

A client running an automotive shop came to me with an AI-built importer. It could not handle the large volume of product data and its many variants. It needed to be fixed to support imports running for many hours.

I reworked the importer to run asynchronously in the background and process records using an iterator. After the changes, it handled imports lasting over ten hours. Processing was deterministic: the same input data and rules produced the same result.

Recommendations

Mateusz carried out an audit and successfully secured a system that was full of holes like Swiss cheese. Everything was completed quickly and in line with the agreed scope.
Grzegorz SłoniewskiMGM S.A.
The security audit was carried out in line with the agreed scope. Based on the report provided by Mateusz, the server configuration was corrected and the identified vulnerabilities were removed.
Radosław Sionek(company name confidential)

Get in touch about an audit.

Describe what the application does, what you built it with and whether anyone is using it. Explain the problem and what you have tried, or what you want checked before launch. I will let you know whether I can help. We will arrange a call to discuss your application and agree on the audit scope.

The data controller is Mateusz Kuźmicz. I use the data submitted through this form to reply to your message and contact you about possible future cooperation. Details are available in the Privacy policy.

I usually reply within 24 hours.