We understand the process first. Then we build the solution.
// 01
The people behind Mandala
Mandala is a software house run by its two founders. We have over 20 years of experience in IT: we design and deliver applications, integrations and AI systems, and we run our own company on the same tools.
- Tomek KupiduraCTO
- Łukasz SomniakCEO
02 / 05 · How we work
How we work
We start by talking to the people who do the work every day and by mapping the data and tools they use. Then we choose the solution: an application, an integration or an automation, with AI where it genuinely helps. After launch we measure the impact and keep improving the system in further iterations.
// process first, technology second
03 / 05 · Company as code
The company as code
We run Mandala from a repository called mandala-as-a-code. Projects, meeting notes, the sales pipeline and procedures are files; decisions and blockers are labelled issues. A COO agent works in the same repo: it writes up meeting notes, turns what was agreed into tasks and prepares a weekly summary.
We run the company the way we build systems for our clients.
04 / 05 · Infrastructure
Our own AI infrastructure
Language models, embedding models, a reranker, speech recognition and speech synthesis run on our own hardware: AI infrastructure built on NVIDIA Blackwell-family GPUs. We process client documents, questions and recordings in-house, without sending them to external APIs. That lets AI work even where data must not leave a controlled environment.
// client data never leaves our infrastructure
05 / 05 · Foundations
Our own foundations
We don't start every project from scratch. We build our own components: agent-core, the base for our specialised agents, and a knowledge base engine with hybrid search and answers that cite their sources. We are also developing our own speech synthesis model. Each of them runs in a real project. The case studies show how.
// Facts
- Experience in IT
- over 20 years
- AI transformation at Expander
- since 2025
- Production knowledge base (22,786 documents)
- over 2 million indexed passages
Technologies we use
// from web apps to AI models on our own GPUs