Ask Hallem-Berg

About

I'm a computer science student at NTNU with a specialization in artificial intelligence, spending 2026/27 on exchange at ETH Zürich. My work spans deep learning fairly broadly: detection and segmentation models for drones and satellite imagery, generative models for synthesizing training data, and retrieval systems built on top of language models.

I build things on my own as well. trafikkskiltene.no started as a way to learn the road signs before my tractor licence at 15, and it still draws traffic and earns advertising revenue years later. veivett.no is the current one, a theory exam prep app covering all four licence classes, currently in beta.

Experience

Netlight

Summer Internship - Data & AI

Jun. 2026 - Aug. 2026

Improved the temporary settlement detection model for SKAI+, the World Food Programme's platform for analysing satellite imagery in humanitarian response. Traced the model's poor generalisation to incomplete annotations rather than model capacity, and built a semi automatic labelling pipeline that gave full coverage of the operating area and roughly ten times as many labels. Adapted YOLO26 with an added P2 head for targets around ten pixels across. The new model found more than twice as many shelters at the default confidence threshold, with higher precision.

Read more →

• Object Detection
• Instance Segmentation
• Ultralytics
• PyTorch
• HPC

Ascend Aerial Robotics Team

Trondheim · Sep. 2023 - Sep. 2025 · Volunteer, 20+ hrs/week

Deputy Chief Engineer

Apr. 2024 - Sep. 2025

Responsible, together with the Chief Engineer, for Ascend's four technical groups, making sure the work of 29 engineers integrated into one drone. Recruited 21 engineers from a pool of over 200 applicants. Sat on the board, setting up the technical budget and handling purchasing for the groups.

• Team Leadership
• Project Management
• System Integration
• Recruitment

Perception Engineer

Sep. 2023 - Sep. 2024

Worked on object detection in the perception group, which turns the drone's cameras, radar and other sensors into what it knows about its surroundings. Trained and fine tuned the detection model on NTNU's HPC cluster, with sliced training through a custom dataloader for targets a few pixels wide. Inference ran on the drone's NVIDIA Jetson Orin, whose throughput set the budget for what the model could look at.

Read more →

• Object Detection
• Python
• PyTorch
• OpenCV
• Ultralytics
• NVIDIA Jetson
• HPC

Computas

Oslo

Summer Internship

Jun. 2025 - Jul. 2025

Developed an AI assistant for NRK TV-Aksjonen, Norway's largest fundraising campaign, in a team of three. Built the backend and Azure infrastructure: SharePoint retrieval through Microsoft Graph connectors, hybrid search in Azure AI Search, serverless APIs in Azure Functions, all deployed with Terraform and GitHub Actions. Answers come with sources attached, and an admin interface lets non-technical staff maintain the FAQs, abbreviations and prompts.

Read more →

• Azure
• RAG
• Python
• TypeScript
• Azure AI Search
• Entra ID
• Terraform
• GitHub Actions

Education

Projects

Veivett

A study tool for the Norwegian theory test for car, motorcycle, moped and tractor. Full syllabus, over 400 explained questions, spaced repetition and realistic mock exams. An ad-free successor to trafikkskiltene.no, built to teach the theory properly, not just test it.

• CI/CD
• Claude Code
• Cloudflare
• Typescript
• Automation

Read more →

veivett.no →

Trafikkskiltene

My first real project, a site for learning the Norwegian traffic signs, built at 15 before my tractor licence. It still gets steady traffic from search and has earned ad revenue every year since launch. Veivett is its successor.

• JavaScript
• HTML
• CSS
• PHP
• Apache

Read more →

751,000

page views

trafikkskiltene.no →