// Software Developer
Sivert Hagelin Benjaminsen
Applied AI & Computer Vision
Stavanger, Norway
Software Developer at Mohn Technology AS, working across disciplines and the full product lifecycle to ship applied AI, computer vision, and backend-driven products. MSc in Applied Computer Science with a Data Science specialisation.
const sivert = { role: "Software Developer", focus: [ "Applied AI", "Computer Vision", "Backend Services" ], location: "Stavanger, Norway", currently: "Mohn Technology AS"};|// About
Background
I am a Software Developer at Mohn Technology AS, building product-oriented software around applied AI and computer vision, backend services, and frontend development.
I work in a small, product-focused engineering team with broad responsibility across the full product lifecycle — from hardware sensors, data capture, and system integration to backend services, frontend GUI development, deployment, and machine learning workflows. Rather than owning isolated parts of a system, I often contribute from early technical exploration through implementation, testing, and real-world use. This has made me comfortable moving between disciplines, seeing how the parts of a product connect, and taking ownership of solutions that have to work in practice.
I have an MSc in ICT Engineering: Applied Computer Science and Engineering with a Data Science specialisation, and a BSc in Information Technology, both from Western Norway University of Applied Sciences (HVL).
// Selected Projects
Work
Production AI/CV — Mohn Technology AS
Biomass Estimation from Fish Images
Worked on biomass estimation from salmon and cod image data. The system uses computer vision and machine learning to estimate fish biomass/weight from images. The pipeline includes segmentation, object detection, image-based feature extraction, and regression models. The salmon biomass model is currently used in practice.
// My contribution
- Worked on segmentation and object detection models for fish image analysis.
- Worked on regression models for estimating fish weight/biomass from image-derived features.
- Built and maintained annotation and data workflows for machine learning datasets.
- Trained and evaluated models using Python-based ML workflows.
- Helped connect model outputs to practical software workflows.
// Architecture
Fish images → Annotation/CVAT → Segmentation/Object Detection → Feature Extraction → Regression Model → Biomass Estimate
Backend Service — Mohn Technology AS
Salmoscan StreamHub Alert Worker
Built a real-time alert pipeline for an aquaculture monitoring platform called Salmoscan StreamHub. The system continuously watches hardware metrics, container health, and application-level errors across a distributed camera and processing setup. It delivers deduplicated notifications to operators through Telegram.
// My contribution
- Built a standalone Go worker that runs alongside the platform's microservices.
- Consumed application error events from a PostgreSQL-backed message queue using PGMQ.
- Queried hardware threshold breaches (CPU, memory, disk) from metrics stored in PostgreSQL.
- Handled container health state changes routed through PostgreSQL triggers.
- Sent Telegram notifications with cooldown-based deduplication to avoid alert storms.
- Used goroutines and Go channels for concurrent processing.
- Implemented context-based graceful shutdown and thread-safe cooldown tracking.
- Worked with Dockerized deployment for x86 and Jetson/ARM targets.
- Added unit tests across threshold logic, cooldown logic, Telegram delivery, and configuration.
// Architecture
App errors + HW metrics + Container health → PostgreSQL/PGMQ → Go alert worker → Cooldown/dedup → Telegram → Operator
Academic Project — DAT255 Deep Learning Engineering
Course-Aware Teaching Assistant with RAG
Built a course-aware teaching assistant for the DAT255 Deep Learning Engineering course. The system generates four types of study support grounded in course material: free-form explanations, multiple-choice quizzes, scored long-answer review, and flashcards. Deployed as a Gradio web app.
// My contribution
- Built a RAG pipeline using a two-pass Markdown splitter, BGE embeddings, and HNSW approximate nearest-neighbour search.
- Implemented a custom decoder-only transformer with GPT-2-small geometry, RoPE embeddings, and KV cache.
- Fine-tuned Qwen2.5-3B-Instruct with QLoRA for improved generation quality.
- Evaluated with Exact Match, Token F1, ROUGE-L, METEOR, BERTScore-F1, and LLM-as-judge scoring.
// Architecture
Course material → Markdown chunking → Embeddings → HNSW retrieval → Context → Generator → Explanation/Quiz/Review/Flashcards
Master's Thesis — Grade A
Real-Time Optical Flow for AUV State Estimation
Master's thesis on real-time optical flow for state estimation in autonomous underwater vehicles. Evaluated Lucas-Kanade, RAFT, PWC-Net, and LiteFlowNet3 with focus on accuracy, computational efficiency, and robustness against visual disturbances. Sensor fusion performed using a Kalman filter combining optical flow and IMU data.
// My contribution
- Evaluated classical and deep learning-based optical flow methods.
- Focused on underwater inspection and visually challenging environments.
- Integrated visual and inertial data through a Kalman filter.
- Thesis received grade A on both written thesis and oral defense.
// Architecture
Camera frames → Optical flow estimation → Feature tracking → Kalman filter ← IMU data → Drift-corrected motion estimate
// Experience
Work History
Mohn Technology AS
Software Developer
- Develop applied AI and computer vision solutions for aquaculture, industrial imaging, and underwater robotics applications.
- Work on biomass estimation from salmon and cod images using segmentation, object detection, regression models, and image-based feature extraction.
- Build backend services in Go and frontend functionality in Vue.js, including APIs and communication between frontend, backend, and worker services.
- Work with Docker, Linux, PostgreSQL, and service-based architectures.
- Collaborate with customers and internal stakeholders to translate product needs into practical AI/software solutions.
HVL Robotics
Trainee / Software Developer
- Worked on a Mixed Reality solution for stroke rehabilitation.
- Conducted user interviews with hospital staff and healthcare stakeholders to understand needs, workflows, and practical requirements.
- Presented project concepts and findings to technical, academic, and healthcare-related audiences.
- Contributed to user-oriented prototyping in an interdisciplinary robotics lab.
// Skills
Tech Stack
> AI / Machine Learning
> Programming / Software
> Tools
// Education
Academic Background
Western Norway University of Applied Sciences (HVL)
Master's Degree — ICT Engineering: Applied Computer Science and Engineering, Data Science Specialisation
// Thesis
Real-Time Optical Flow for State Estimation in Autonomous Underwater Vehicles
Grade: A
Western Norway University of Applied Sciences (HVL)
Bachelor's Degree — Information Technology
// Thesis
Robot Arm Trajectory Planning using Mixed Reality
Grade: A
// Interests
Outside Work
Outside of work I spend most of my time outdoors and staying active.
Rock Climbing
Problem-solving and persistence, one route at a time.
Skiing
Making the most of the Norwegian winters.
Volleyball
Team play and staying active.
// Contact
Get in Touch
Feel free to reach out via email or connect on LinkedIn.
// Built with Next.js, Tailwind CSS, and TypeScript
// © 2026 Sivert Hagelin Benjaminsen