AI & Machine Learning
Engineer

Ongoing computational research at UiT Narvik’s BEaM group — applying machine learning and data engineering to support sustainable energy and environmental engineering studies.

• Build data-driven workflows for faculty-led research: literature synthesis, structured datasets, and predictive modeling
• Apply Python-based ML and statistical methods in a cross-disciplinary research setting
• Collaborate with supervisors on analysis, documentation, and academic outputs

Delivered production features for a multi-tenant B2B SaaS platform — same governance patterns applied in industrial AI contexts where consequential decisions require human approval gates.

• Built production features on a multi-tenant B2B SaaS (NestJS, Next.js 15, TypeScript): RBAC, per-tenant data isolation, notification workflows, and document flows with traceability requirements
• Shipped a Gemini AI-assisted drafting feature with a human approval gate before any AI-generated content was published — same governance pattern required when AI operates near consequential decisions
• PostgreSQL schema design, Prisma migrations, encrypted asset storage on AWS S3; GitHub Actions CI/CD introduced from scratch
• Code review and agile sprints with stakeholder demos

Full-time embedded at HHS Robotics, building AI systems on live production data from an operational warehouse robot. Research conducted on operational systems and published as M.Sc. thesis (UiT, 2025, NVA).

• ARMADA (anomaly detection): Context-Augmented Anomaly Detection combining Isolation Forest, One-Class SVM, and LOF with fine-tuned GPT for maintenance recommendations. 43.5% improvement in detection precision, 13.7% false-positive reduction, and 97.1% faster training than Anomaly Transformer baseline on live telemetry data
• FACTS (document intelligence): Layout-aware PDF processing with LLM-based factuality verification. 98% factuality accuracy vs 53% TF-IDF baseline; 37.2% improvement in information extraction accuracy and 29.8% reduction in verification errors across factual, procedural, list, and reasoning query types
• Kafka-to-PostgreSQL real-time telemetry pipeline; FastAPI services; Streamlit operational dashboards; all results statistically validated (Welch's t-test, formal error bounds)

Led shift operations for a team of 8–12 people, developing practical judgment for when to resolve independently and when to escalate — a pattern that transfers directly to cross-functional technical work.

• Task scheduling, stock checks, and real-time coordination to maintain service quality under high-volume periods; escalated equipment and supply issues to management with clear situational summaries so decisions could be made quickly without losing context
• Handled customer conflict resolution directly and on the spot, de-escalating situations calmly while keeping the rest of the team focused on operations