AI Planet: AI that truly speaks business language

[This article is part of a content series developed in collaboration with FEDIL, showcasing how artificial intelligence is contributing to the digital transformation of Luxembourg’s economy.]

The figures are unequivocal: whilst companies are investing heavily in artificial intelligence, the vast majority of projects fail to generate genuine business value. The reasons are manifold. "Today's major language models are designed to be generalist. What they fundamentally lack is specialisation—the ability to understand and speak the language of a specific company, industry or community," explains Chanukya Patnaik, Founder & CEO of AI Planet, a deeptech AI startup dedicated to educating and democratising artificial intelligence for all.

The problem is primarily structural. Out-of-the-box AI solutions lack familiarity with compliance terminology, institutional processes and the regulatory environment unique to each organisation. The result: hallucinations, security breaches, complex integration and, crucially, no measurable return on investment. Companies consequently accumulate disconnected AI experiences, devoid of intelligence or institutional memory.

An AI brain that learns your craft

In response to this challenge, a new approach is emerging: the institutional AI brain. Rather than multiplying generic tools, the focus shifts to building a centralised, secure and specialised AI infrastructure capable of reasoning within each organisation's specific domain—be it banking, legal, healthcare or public administration.

This approach rests on three pillars:

• A secure AI marketplace that ensures privacy and compliance
• An advanced orchestration engine enabling the deployment of intelligent AI agents that automate complex workflows
• An open-source architecture promoting continuous evolution and autonomous learning

"Over time, this ecosystem evolves into a village of AI agents, where specialised agents collaborate, share memories and learn continuously, eventually leading to a collective intelligence. This collective intelligence forms a living institutional or enterprise AI brain that aligns deeply with their processes and goals," Patnaik notes.

With the support of MeluXina

The Luxembourg example perfectly illustrates this need for specialisation. Although Luxembourgish is a national language, it remains largely under-supported by mainstream AI models. Consequently, generalist models achieve poor performance scores on basic Luxembourgish tasks. Building upon its Open Source LLMs and Agentic OS platform, AI Planet developed LuxLlama—a model specifically fine-tuned on 400,000 examples combining Luxembourgish language teaching data with reasoning tasks relevant to the linguistic context. AI Planet achieves a performance score of 74.6, compared to 61.4 for the best generalist model.

Simultaneously, the company developed LUXELLA, the first comprehensive benchmark specifically designed to assess LLM capabilities in Luxembourgish. It comprises 15 distinct language categories across four difficulty levels, using a consistent judging system based on existing LLMs. This project received support from the Luxembourg government, with computing resources allocated by MeluXina. The first model version was released in spring 2025.

Together, LuxLlama and LUXELLA have the potential to enable Luxembourg to deploy its own sovereign, language-sensitive AI infrastructure, built securely on its own terms.

Tenfold savings

The gains are tangible and quantifiable. Companies deploying an institutional AI brain realise up to tenfold cost savings through workflow personalisation and automation. Revenue increases reach 25 per cent for customers using advanced workflows driven by AI agents.

Operational efficiency improves by 30 per cent in manufacturing, banking and other industries, thanks to intelligent agents that genuinely understand business context. These results stem from mass adoption: over 200,000 downloads of open-source models and a global community of 300,000 members across more than 150 countries.

Numerous developments ahead

AI Planet continues refining and extending LuxLlama with broader datasets and domain coverage to deepen Luxembourgish language mastery.

"We hope to develop chatbots and citizen service agents fluent in Luxembourgish, tools to assist with machine translation and grammar, as well as legal, medical and financial support for workflows where linguistic accuracy is crucial," Mr Patnaik explains.

By expanding Agentic OS adoption in regulated industries—banking, healthcare, pharmaceuticals and government—where security, privacy, compliance and governance are non-negotiable, AI Planet aims to enable every company to build and operate its own secure institutional brain.

"We aim to continue our open-source contributions. By releasing new models, frameworks and benchmarks to the community, we want to strengthen our position at the forefront of enterprise research and deployment," Mr Patnaik concludes.