BSc student in Machine Learning and Artificial Intelligence
at the University of Oslo, graduating 2028.

| LinkedIn | andgron@uio.no |

Punktlig — Machine learning models that beat Entur's official public transport delay predictions.

Entur publishes realtime arrival estimates but no history, so I archive both the predictions and the ground truth and train against the live production system as the baseline. Gradient boosted trees on a time-based split with no lookahead: 23% lower MAE than Entur's own estimates, weighted across horizons, on 167k held-out rows. Quantile regression adds calibrated arrival intervals, which the official feed does not offer at all.

The README documents the ablations, the learning curve, and the cases where the model loses.

Tenki Labs — Cofounder, 2026. Norwegian AI lab training and serving local LLMs on hardware it owns, with open benchmarks for Norwegian. I was the lead developer and worked on Hugin and Munin, the Norwegian fine-tunes, and on the research around them. Sold and exited in 2026.

LMC Simulator — A Little Man Computer assembler and executor in Python, with branch labels, ASCII output and a library of example programs. Built because the browser-based simulator we were recommended in IN1020 was old, couldn't save anything, and I wanted something that ran in the terminal.

Technologies I have used extensively

Technologies that I have used, but not extensively

Technologies I'm picking up now