Arca — applied AI studio · Norway

AI systems that read your documents, so your engineers don’t have to.

I turn decades of scattered files and manual processes into structured, searchable, automated systems — built for industrial companies that need working software, not demos.

Start a project See the work ↓

arca (lat.) — chest, vault; the root of “archive.”

01 — About

I’m Nicholas Johnsen. Arca is my studio.

I’m studying electrical engineering at NTNU in Trondheim, with a passion for AI engineering — and real experience shipping it in industry. I work on a specific kind of problem: decades of knowledge trapped in documents, and skilled people spending their days moving information around by hand.

I work across applied AI — document intelligence, RAG, agentic automation — with a bias for systems that run where your data lives. When confidentiality matters, I build on self-hosted models: your documents never leave your infrastructure.

I’m not an agency. When you work with Arca, you work with the engineer who scopes the problem, builds the system, ships it — and answers for it afterwards.

Based in
Trondheim, Norway
Studies
Electrical engineering · NTNU
Focus
Applied AI for industry
Methods
Document intelligence · RAG · agents
Privacy
Local & self-hosted models
Clients
Industrial & engineering companies
02 — Work
NOV logo

Client — NOV · Global energy equipment

At NOV, a global oil & gas equipment company, I’ve delivered two internal systems that turn decades of legacy engineering files into tools engineers actually use. Open each project for the full story.

02.1 Smart document library Decades of legacy files — structured, searchable, found in seconds.

NOV’s engineering knowledge lived in an aging document system: decades of files, findable only if you already knew where to look. I built a smart library system that ingests and structures the archive and puts fast, smart search on top — so the right document is seconds away instead of an afternoon away.

Deliberately boring engineering: the library runs on plain, deterministic Python — no AI at runtime to drift, hallucinate or need a GPU. AI agents were used heavily to build the system, not to run it. The right tool for the job.

  • One structured library replacing a clunky legacy system
  • Smart search across decades of files
  • Deterministic and dependable — no black box in the critical path
02.2 AI tendering agent Reads HMBs, PFDs and drawings — fills component datasheets automatically.

Tendering meant engineers reading heat & material balances, process flow diagrams and engineering drawings, then re-typing values into component datasheets — by hand, for every bid. I built an AI agent that does the reading and the filling: it extracts the values and auto-populates the standardized datasheet template for each component. Engineers review the result instead of producing it.

  • Datasheet population: hours of manual entry → minutes of review
  • More consistent tenders, fewer transcription errors
  • Fills the standardized templates engineers already know
02.3 Your project This slot is reserved for what we build next.

If you have decades of files nobody can search, paperwork your engineers do by hand, or an AI idea that needs an engineer — this space is waiting for your story.

Start the conversation →

03 — What we can build for you
  • 03.1

    Document intelligence & smart search

    Decades of legacy files become one searchable system — inside PDFs, drawings and scans, not just filenames. Ask a question; get the page.

  • 03.2

    AI process automation

    Repetitive engineering paperwork handled by an agent: datasheet population, form filling, tender preparation. Your people review — they stop retyping.

  • 03.3

    Private RAG & self-hosted AI

    AI on your data, under your rules. On-premise or private-cloud systems for confidential documents — nothing leaves your infrastructure.

  • 03.4

    AI consulting & implementation

    From “where would AI actually help us?” to a shipped system. Scoping, prototyping, delivery — one engineer, accountable end to end.

Every engagement starts the same way: a short call about the problem, then a concrete proposal. Tell me what’s slow.

04 — Contact

Tell me what’s slow.

A document archive nobody can search. A process your engineers do by hand. An AI idea you want a second opinion on. Send a few lines — no deck required.

Prefer email? Write to nicholas.johnsen@outlook.com

Usually replies within one working day