Things I built to see if they would work.
// agents, robots, scrapers, and a few weird experiments
> Some became products. Some became research. Some taught a robot to move or an agent to play LUDO. All of them forced the idea to meet code, data, and constraints.
// Selected Work
Models meet code, sensors, and bad assumptions
> Industrial robots, scraping workers, game agents, and models that listen. The useful part starts when the clean notebook ends.
[Zero-Cost Job Scraping at 500k Listings]
FastAPI workers, isolated subprocesses, and Postgres as the queue. 500k listings ingested on self-hosted infra at €0.
[Improvements in throat swabbing key-points detection]
A smaller PointNet pipeline for detecting key points inside the mouth from 3D point clouds. Built with Lifeline Robotics and SDU.
[Visual Navigation using Deep Reinforcement Learning]
An A3C agent learning to navigate homes from pixels inside AI2-THOR simulations.
[LSTM Time Series Prediction for RAS Systems]
LSTM forecasts for salmon biomass using telemetry from Billund Aquaculture recirculating systems.
[AI Learns to Play LUDO]
A Deep Q-Learning agent that reached a 92% win rate against random LUDO opponents.
[Speaker Recognition using Deep Learning]
A compact CNN that identifies podcast speakers from five seconds of audio and a small dataset.
[AI Learns to Play Flappy Bird]
An early NEAT experiment that evolved Flappy Bird controllers without hard-coded rules.
[Elsewhere]
> A hackathon win, a talent program, and helping new students survive Denmark.
Future Talent Denmark
A six-month program connecting international engineers with Danish companies.
> Learn more
1st Place SDU Hackathon 2021 sponsored by Danfoss
Led the winning design for an autonomous free-range egg harvester. We took home 15,000 DKK.
> Learn more
SDU - International Buddy
Helped new international students figure out university and life in Denmark.
> Learn more