Append Consulting
Eighteen months as an ML engineer at Append, a small AI consultancy in Oslo. Most of the work was shipping production machine learning for Norwegian public-sector and industrial clients, the kind of projects where someone real depends on the model working tomorrow morning. It was also where I learnt that "the model" is usually the smallest part of the job.
What I shipped
Four production projects over the period. Three were for clients under NDA, so I cannot describe their details. The fourth is its own deep-dive.
- Helsedir RAG. A retrieval-augmented question-answering service for the Norwegian Directorate of Health. (client confidential)
- Digdir analysis. Embedding clustering and LLM summarisation over a large corpus of public-innovation documents. (client confidential)
- Hydro Aluminium. Optimisation and ML for an industrial process at a Norwegian aluminium producer. (client confidential)
- ML competition. Designed and ran an internal Kaggle-style competition between NTNU students and Hydro engineers, as a way of seeding more ideas around that engagement.
What I learned about consulting
The hard part is rarely the model. It is the data you cannot get cleanly, the stakeholder who needs the explanation in their own vocabulary, the production-environment quirk that nobody flagged until the day of deployment, and the surprisingly long path from a notebook that works to a service that someone else can operate without you. I leave Append more boring than I came in, and I consider that a win.
Why I left
I joined Infinigrid in January 2026 to work on grid-scale ML full-time. Consulting taught me a great deal; I wanted to go deep on one problem rather than wide across many.