Haukur
Sigurður Haukur Birgisson
Haukur, Icelandic for hawk.
In primary school, everyone received a micro:bit, a tiny programmable computer meant to teach basic coding. Most of my classmates lost interest after a week. I started collecting their discarded ones, taking them home, wiring them together, making them talk to each other. I didn't know what I was building. I just knew that when you have limited resources, you learn to make things work.
That instinct led me to a problem most of the tech world ignores: Icelandic. A language spoken by roughly 370,000 people. Too small for Big Tech to care about, too complex for off-the-shelf solutions. When I started working on OCR and text-to-speech systems at Miðeind, I discovered that the tools I needed didn't exist. So I learned to build them from scratch, training models on synthetic data I created myself, finding solutions in obscure papers that mainstream research had forgotten.
Now I'm working towards my AI degree at University of Groningen, still chasing the same question that hooked me as a kid: what can you build when no one's built the pieces for you?
I compete internationally in cybersecurity with Iceland's national team. I sail competitively for Iceland. And I keep shipping systems for a language the world overlooks — because someone has to, and I've been training for this since I was collecting micro:bits from kids who didn't see what I saw.
Curious to learn more?
Read more, Contact meHugleiðingar
Literal translation is Roads of Mind —pick a path!
How to Install R on Mac - Complete Setup Guide
Step-by-step guide to install R and set up tooling on macOS for data science.
RjochtWurd: Empowering Frisian Through Speech-to-Text Improvement
Building RjochtWurd to boost Frisian speech-to-text quality with corrective models.
Is it Possible to Distinguish Between AI and Human-Generated Text Using Watermarking Techniques?
Exploring watermarking approaches for detecting AI-generated text and their limits.
How do we make great language models for smaller languages such as Icelandic?
Tokenizer-first approach to training high-quality Icelandic language models.
Predicting footfall in every pool in the capital area in real-time with ML
Real-time pool footfall predictions in Reykjavík using weather-aware ML models.
From Snorri to RNNs
Tracing Icelandic literary heritage to modern recurrent neural network applications.
Predictive Modeling for the Water Industry in Iceland
Forecasting water demand and operations for Iceland's utilities with machine learning.
Forecasting Car Accidents in Iceland and the US: A Machine Learning Approach
Comparing accident risk models across Iceland and the US with machine learning.
Icelandic word completion with stateful RNN models
Building Icelandic word completion models with stateful recurrent networks.
Multi-class text classification for the Icelandic language
Training multi-class classifiers for Icelandic text with modern NLP techniques.
Verkefni
A showcase of my selected projects
Reach Out
Want to get in touch? Feel free to reach out to me at the following:
- Email: contact@sigurdurhaukur.com
- LinkedIn: Sigurður Haukur Birgisson
- For more details visit the contact page.