[X]
[LI]
[GH]
[MAIL]
[Projects & Experiments]

Building reliable AI that works in the real world.

Robotics & ControlVision & AudioReinforcement Learning

> A curated slice of research, prototypes, and playful experiments I have built across robotics, machine learning, and human-centric AI. Each project pairs rigorous engineering with an obsession for smooth, resilient user experiences.

// Selected Work

Applied AI across hardware, research, and product

> From industrial robots to conversational agents, these builds explore how ML models perform once they leave the lab and meet physical constraints, noisy sensors, and real users.

Zero-Cost Job Scraping at 500k Listings
System DesignScrapingPostgres

[Zero-Cost Job Scraping at 500k Listings]

Zero-cost scraping fleet with FastAPI, subprocess isolation, and Postgres-as-queue to ingest 500k listings without queues or Kubernetes.

Improvements in throat swabbing key-points detection
3D VisionRoboticsMedical

[Improvements in throat swabbing key-points detection]

Enhanced 3D point-cloud perception for oral cavity swabbing with a modified PointNet pipeline co-developed with Lifeline Robotics and the University of Southern Denmark.

Visual Navigation using Deep Reinforcement Learning
Reinforcement LearningSimulationNavigation

[Visual Navigation using Deep Reinforcement Learning]

A3C-based visual navigation for home robotics, training embodied agents on photorealistic AI2-THOR environments to generalize across diverse layouts.

Adaptable Impedance Learning for UR5 Robotic Manipulator
ControlGMM/GMRRobotics

[Adaptable Impedance Learning for UR5 Robotic Manipulator]

Learning-from-demonstration system combining GMM + GMR for force-based variable impedance, enabling safer, adaptive UR5 manipulation.

LSTM Time Series Prediction for RAS Systems
Time SeriesAquacultureLSTM

[LSTM Time Series Prediction for RAS Systems]

Forecasting total salmon biomass in recirculating aquaculture systems using LSTM networks and production telemetry for Billund Aquaculture.

AI Learns to Play LUDO
Reinforcement LearningGamesDQN

[AI Learns to Play LUDO]

Deep Q-Learning agent that masters the LUDO board game, reaching a 92% win rate against stochastic opponents after curriculum-style training.

Speaker Recognition using Deep Learning
AudioCNNBiometrics

[Speaker Recognition using Deep Learning]

Compact CNN architecture for speaker identification from five-second podcast snippets, optimized for low-data, low-latency scenarios.

AI Learns to Play Flappy Bird
NeuroevolutionGamesExperiment

[AI Learns to Play Flappy Bird]

An early experiment with the NEAT algorithm, evolving controllers that consistently clear Flappy Bird obstacles without handcrafted heuristics.

[Beyond the code editor]

> Community, competitions, and mentorship keep me grounded in the people who use the things I build. A few highlights:

Future Talent Denmark

Future Talent Denmark

Six-month bridge program connecting Danish firms with international talent and accelerating integration into the Danish tech ecosystem.

> Learn more
1st Place SDU Hackathon 2021 sponsored by Danfoss

1st Place SDU Hackathon 2021 sponsored by Danfoss

Led the winning concept for a sustainable free-range farm egg harvester, earning top honors and a 15,000 DKK prize.

> Learn more
SDU - International Buddy

SDU - International Buddy

Volunteer support for new international students adapting to life in Denmark, guiding their academic and cultural onboarding.

> Learn more