AI-assisted software development & engineering.Independent evaluation of AI-generated code.

Describe what you want to build, fix or check. An engineer replies within one business day.

Talk to an engineer Explore AI Institute programs

PRs merged to main in this site's repo228

DNV Management System Certification ISO 9001

ISO 9001 certified (DNV)
Training eligible for BUR funding
Engineers certified by Anthropic
Member of the West Pomeranian ICT Cluster
In business since 2012

Evaluating AI-assisted code

This is what evaluating AI-assisted code looks like.

An evaluation starts with a sample of files or pull requests from your repository: each one gets a score against four criteria: correctness, security, maintainability and test quality. The result is a report with findings ranked by severity - like the one beside this text - and a walkthrough with your team. The service runs as a one-off or as continuous review, during the work and after release. This is not a tool: an engineer reads and evaluates the code, and findings land in your issue tracker, not in another dashboard.

Evaluation is one of the service areas at 23Minds ↓

users_api.pyAI-generated

01def get_user(req):
02    uid = req.args.get("id")
03    row = db.query(
04      f"SELECT * FROM users WHERE id={uid}")
05    tok = jwt.decode(t, verify=False)
06    return row

●Unsanitized SQL from request inputL3-4

▲JWT signature verification disabledL5

✓No secrets committed in source-

3 findings · 1 criticalEval · v2.4

AI training for leadership, product teams and engineering.

Programs for business, product and engineering teams: workshops priced per day, or a cohort program running over several weeks. The curriculum is built around your team's tools and the work they actually do. Exercises run on the team's own repositories and processes, not on generic AI examples.

Leadership and compliance

AI Governance and EU AI Act Essentials

System classification, human oversight, rules for using AI.

Product teams

Designing AI Use Cases That Reach Production

From an idea to a pilot with an end date and a decision.

Engineering

Secure Coding with AI Assistants

Mistakes typical of assistant-written code, agent permissions, review rules.

Engineering

AI-Assisted Development with Coding Assistants

Claude Code, Cursor and Copilot on your team's repository.

All teams

AI Literacy for Business Teams

Prompts for real work tasks, data you must not send, checking the answers.

Engineering, QA

Reviewing AI-Generated Code

What to check in a large AI-written change, and what blocks a merge.

Explore Institute programs View training catalog

Services

From code to production.

Build & ModernizeCustom Software Development with AIAgents write the code, and every change passes tests and two reviews.Evaluate & SecureIndependent AI Code AuditA code sample scored on four dimensions, against published criteria.Build & ModernizeProcess AutomationThe automation stays with you as a procedure with an owner and a change log.Evaluate & SecureSecurity ReviewSecrets and vulnerabilities in your code, with steps to reproduce and a fix.Build & ModernizeLegacy ModernizationReplaced component by component, with a way back for each.Operate & MeasureLLM ObservabilitySee model calls, errors, latency and cost.

More services

All services

What you'll see, and when

A working rhythm you can plan around.

Week 1
Scope in writing
An intro call and an agreed scope: what you get, in what order, at what price. No commitment on your side.
Month 1
The first tangible result
The first evaluation report, or the first working slice of the system on a preview. Findings land in your issue tracker.
Quarter 1
A rhythm instead of pushes
Continuous review, retests after fixes, quality gates in your process. The team works with AI by agreed rules, not by feel.
Year 1
The capability stays with you
Procedures, evaluation criteria and trained people are yours. 23Minds stays for the hard cases and for whatever changes in the tools.

Timelines are indicative and depend on scope; dates are set on the intro call.

FAQ

Questions procurement teams ask.

We start with an intro call to map your situation, then scope a fixed first engagement - an evaluation, a workshop, or an automation pilot - with clear deliverables and timeline.

Every practice has an entry package with its price on the practice page: from a training day to an audit of one repository. Where the effort depends on the size and complexity of your system, the price is shown as "from" and becomes fixed once the scope is agreed on the intro call, with no hidden costs. Larger projects and retainers are quoted after the call, per phase.

Selected AI training programs are listed in Poland's Development Services Database (BUR); the operator decides on co-funding and its level, based on company size and region. We check it with you on the intro conversation.

Two things. We are engineering-led, so you talk to the people who write and maintain the code rather than to a layer of account management. And we leave the capability with you: we train your team to work with AI instead of tying it to us permanently. Outcomes are measured in production, against indicators agreed before we start.

Design and development of web and mobile applications, e-commerce and websites; application service; IT training and consulting; online marketing; and robotic automation of business processes.

Talk to an engineer.

Describe what you want to build, fix or check. A reply within one business day.

Talk to an engineer

Szczecin - ul. Wawrzyniaka 6WWarszawaZielona GóraKraków