The AI code review that catches real bugs
From auth bypass to deadlock: what our AI code review caught before the human review, on publicly visible Pull Requests.
The PRs on GitHub →PlaygroundNotes from the engine room
PlaygroundWhat we learnedabout production AI,written down.
Writing from the engine room: how we take AI from pilot to live operation, what breaks, and what holds up. Not trend posts, but experience. Our public systems run live, and our code is open on GitHub.
Notes
What we learned in the engine room
Production AI in practice, alongside our public systems, where it can be checked. The ones marked "Live" can be seen running or in the source code.
Multi-LLM orchestration in live operation
Fallback chains, cost control and duplicate filtering in a fully autonomous pipeline: on the order of a hundred articles a day, with no human intervention.
news.lavx.hu →RAG that names its source
How an internal knowledge hub is built that does not hallucinate but cites, and gives back engineer-hours on the order of a week.
The system on the Proof page →The data stays where it has to
From on-prem to EU region: what data residency means in practice, and what the GDPR and the EU AI Act actually do not require.
The Trust Center →Benchmarking models, automatically
How we trial new LLMs through our multi-provider gateway, and what decides which model fits the task.
tools.lavx.hu →The Playground is growing: the systems above run live, and the detailed write-ups go out continuously. Want to hear about new ones? Write to us and we will let you know.
Let's talk
Working on a similar task?
If any of the above sounds familiar, let's talk it through. Within one working day an engineer replies, not a salesperson.