Overview
Michiel Van Beirendonck is the CEO and co-founder of Belfort, based in Leuven. He co-founded the company in July 2024, where he has also served as chief technology officer.
From October 2019 to August 2025 he was a PhD researcher at KU Leuven. During that period he held silicon security internships at Rambus from July to October 2022 and at Microsoft from May to August 2023. In the summer of 2018 he was a student researcher at EPFL (École polytechnique fédérale de Lausanne).
He studied electrical and electronics engineering at KU Leuven, completing a BSc between 2014 and 2017 and an MSc between 2017 and 2019, and spent 2017 to 2018 at EPFL on a SEMP exchange.
Career history
In the news
- A few weeks ago, at the Syndicate One Summit panel, I was asked what first-time technical founders need from their backers. My answer: a strong ecosystem, and the chance to learn from other founders. Christine Hick then uno-reversed the question and asked whether I’ve taught other founders. As an early founder I haven’t had many chances yet, but I’m grabbing every one I can! So next week I’ll be at KU Leuven KICK House for “The Co-founders Conversation You Need to Have,” together with Sam Biermans from Nunki.ai. We’ll cover the
- Best support ticket I ever filed. 📩 Meeting and interviewing Samuel Berton at Tectonic turned out to be one of the luckiest hires we’ve made. Now we’re looking for the next people to join us, and work directly with Samuel Berton, me, Laurens De Poorter and the rest of the team: https://lnkd.in/e3T-fDjr
- Last week we launched Belfort eML, our encrypted machine learning cloud. First milestone down, many more ahead. Making encrypted computing fast and usable has always been at the core of what we do, and none of that happens without our incredible team. Every week someone figures out how to support more models that run encrypted end-to-end, or how to squeeze larger models onto the same hardware stack. But now our hardest problems are shifting. We need to decide which products to build next, which AI models to make private first, and
- During my PhD, my focus was always on making encrypted computing fast. At Belfort, we’re now just as focused on making it accessible. When we started, speed was the biggest hurdle to adopting Fully Homomorphic Encryption. But in our early customer conversations, a different barrier kept coming up: you needed to be a cryptographer to use it. So today we’re launching Belfort eML, our encrypted machine learning cloud. You write PyTorch, and that’s it: ✅ No fiddling with cryptographic parameters ✅ No rewriting your network into
- My first panel ever, and I didn’t even have to leave our office building. 🎤 Last Friday, Syndicate One hosted its 10th Summit at De Hoorn in Leuven, with the theme “Small Funds, Big Impact”. Christine Hick asked Maxime Lahy and me what first-time technical founders need from their backers. I went straight from a PhD at KU Leuven to co-founding Belfort. Within a year, we grew to a team of 18, partnered with companies like Google and LG Electronics, and learned how to build a company across Leuven and San Francisco. A PhD prepares
- Making encrypted computing fast and accessible has always been our mission. Today, we're also making it fun. 🎙️🎲 We just launched a game you play with your voice, where the AI listening never actually hears what you say. You get 12 hints and 20 seconds to guess the hidden words. Our hotword model checks your guesses, but your audio is encrypted before it leaves your device. The server runs the model directly on the encrypted audio, and only you can see the result. Why does this matter? Voice assistants have a track record of
- My first opinion article in the Belgian newspapers! Belgium is world-class in cryptography, now let's build a sector around it. A few weeks ago, I wrote about the quantum threat and the urgent migration to post-quantum cryptography by 2030. That story continues. Belgium has long been world-class at research in exactly this topic: cryptography and data security. What's often been missing is our own industry of early adopters to capitalize on it. Shifting that balance is one of the core motivations why we founded Belfort. Data security
- Today we’re launching Belfort’s first showcase of encrypted ML: classification of encrypted images. 📸 This is running ResNet-20, fully under encryption, in 163ms. The fastest out there, 3x faster than any previous state of the art. 🔥 There’s no accuracy tradeoff: 92.5% encrypted vs. 92.8% in plaintext. The model doesn't notice it's encrypted. This is just the first demo of what's coming with our upcoming Cyclops release. Try it yourself at https://lnkd.in/dpBG69vY, and stay tuned for more! If you're running AI workloads and
Related profiles
This page shows public professional information only, each fact cited. Is this you? send a correction, or ask for removal within 24 hours, no questions asked.