Luxembourg works in several languages, and braddl wants its AI tools to do the same. The startup’s braddl Notes turns multilingual meetings into minutes, powered by its own Luxembourgish speech models.
Can you pitch your company in a few words?
braddl builds AI that understands Luxembourgish the way people actually speak it, mixed with English, French and German in the same conversation.
We develop and train our own speech and language models. Our first product, braddl Notes, turns meetings into transcripts and structured minutes.
Why did you decide to create this company?
AI was automating everyday work everywhere, but Luxembourgish was left behind: inaccurate transcripts, unnatural pronunciation and systems that drifted into German. Companies here kept telling us they could not find tools that work in their multilingual environment.
Daniel founded AI Lime, whose Lexplain brings AI legal research to Luxembourg and European law. Building for lawyers taught him to handle strictly confidential information, from retrieval and security to compliance and model evaluation, and to run a product for demanding professionals.
Leminh was a core contributor to Deepgram’s Nova-2 speech recognition model in Silicon Valley. At its release it outperformed OpenAI’s Whisper, and it has since reached millions of users through note-taking apps such as Granola.
Together, we are building the tools Luxembourgish was missing.
What are your product(s)/service(s) and/or your business model?
braddl Notes joins Microsoft Teams and Google Meet, or records around the table. Unlike general note takers, it follows meetings where people switch between English, French, German and Luxembourgish, tells the speakers apart and produces structured minutes.
Chatbots powered by our language model answer from an organisation’s own documents and knowledge bases, in English, French, German or Luxembourgish, whichever language the citizen or customer writes in.
Voice agents combine our speech recognition, language model and speech synthesis for spoken interactions in Luxembourg’s languages.
We serve businesses and public organisations through product subscriptions, and license our models for integration into other systems.
How is your offering different from your competitors and other market players?
Measured on the same Luxembourgish audio, our speech recognition makes 36% fewer errors than the best public Luxembourgish system and three times fewer than global tools attempting the language.
We own every layer: model development, training, deployment and the applications on top. So when something goes wrong in Luxembourgish, we improve the whole chain instead of patching around it, and customers get better results.
Our AI runs in Luxembourg: recordings, transcripts and minutes are stored here, and transcription and summarisation run on our own infrastructure. For organisations handling confidential information, that matters as much as accuracy.
Why did you choose to join the Luxembourg-City Incubator / House of Startups?
The Luxembourg-City Incubator supported us with our previous companies, so incubating braddl there again was a natural choice. It puts us among people who understand what starting a company takes, just as we turn our technology into a business and win our first customers.
What’s next?
braddl Notes is in a running pilot with a large municipality in Luxembourg, our pilot partner. We are proving it in everyday meetings and measuring the time it saves.
Next, we bring our chatbots and voice agents to market, built on the same speech and language models.
We welcome conversations with organisations that want to pilot braddl Notes or bring Luxembourgish AI into their own workflows.
Who are your main users or clients, and how do you engage with them?
We work with municipalities, public administrations and businesses whose teams operate in several languages. For braddl Notes, the users are the people who run meetings and write them up.
We sit down with users and decision-makers to see how minutes are produced, reviewed and shared today.
Pilots show us transcript quality and how much correction the minutes still need. That feedback shapes both the application and the models underneath.
How do you explain your value to someone outside your industry?
Imagine a meeting where colleagues speak English, French and Luxembourgish. Afterwards, someone has to turn that conversation into a clear written record.
braddl Notes writes the first draft with the key points, decisions and actions, ready for the team to review. People concentrate on the discussion instead of taking notes.
What’s one hard truth you’ve faced since launching?
The hard truth: real Luxembourgish meetings are far messier than any demo. A speaker starts a sentence in Luxembourgish, quotes a French regulation and finishes with an English technical term, while someone else talks over them. Local names, accents and background noise come on top.
Being dependable in exactly those meetings is the bar we set ourselves. Every improvement to our data, training and evaluation is aimed at it, and we measure success by the minutes people actually rely on.
What do you hope your startup will change in the next year?
Over the next year, we want braddl Notes to become a regular working tool that cuts the time spent writing minutes and makes meeting decisions easier to follow up.
That is what organisations here keep asking for: AI that handles Luxembourgish well, which they struggle to find.
Luxembourg works in several languages; its AI tools should too.
