Kapllan Labs exists to build enduring AI capability in the Nordics.

K—1K—SpeechK—VoiceK—ListenLlanaPllan

We focus on building the knowledge required to develop advanced AI capability over time, close to our own languages and institutions.

Oslo · StockholmResearch · Engineering · ApplicationEst. 2026

We are not building toward a single app or a single model. We are building the knowledge, methods, technical competence and systems understanding required to develop advanced AI capability over time.

Capability compounds. Products are what it looks like on a given day.

We believe critical AI competence should not only be rented from external systems. It must also be developed locally, close to our languages, institutions, industries, values and real world problems.

Proximity is not sentiment. It is what makes a system correct in the place it is used.

Kapllan AI is a lab where research, engineering and practical application come together. The systems, models, tools and products that emerge are the outcome of that process working.

One building, one loop: read, train, ship, learn, repeat.
What we offer

Five ways to work with the lab. All of them leave capability behind.

Applied research partnership

Bring us a hard problem. Keep the method.

A real problem from your organisation, run as a research engagement: scoped, measured, written up. You keep the system and the reasoning behind it.

Engagements run in quarters, not sprints
Domain models

Fluent is not the same as correct.

Models adapted to one field, on your data, in your language, held to your rules of evidence.

Trained and evaluated in house
K—Speech

Nordic speech, built from the ground up.

Voice and transcription for the Nordic languages, with the evaluation that proves they hold in the room.

Contact centres · clinics · field operations
Agentic workflows

Systems that ask before they guess.

Reading, planning the next step, escalating when unsure, inside the processes you already run.

Claims · casework · back office
Capability transfer

We would rather be unnecessary.

Your team trained alongside ours and the system handed over running, so the competence stays in the building.

For teams that intend to own their AI stack
In the lab

We train in house, and we watch the numbers.

runs.kapllanlabs.io / k-legal-1 / model metrics134k steps · 2 epochs · live
The naming

Where the names come from, drawn out loud.

The stack

From the rack to the review console, we run the whole pipeline ourselves.

Metal

Enterprise racks, GPU nodes and storage we own and operate, for everything that runs day to day.

Dell VxRailiDRAC9TrueNASNVIDIA

Frontier compute

Frontier clusters back the stack when a run needs more than the house can give.

GB300GB200H200burst clusters

Cluster

Virtualisation, networking and scheduling across the nodes, one addressable cluster.

Proxmox VERayDockerSpark

Data

Collection, cleaning, dedup and versioning of the corpora we train on.

ingest pipelinesdedupdataset registry

Training

Pretraining and post training runs, tracked end to end with every checkpoint kept.

MLflowcheckpoint storerun log

Evaluation and review

Automated evals plus our own human review console, where native speakers judge the output.

eval harnessK—Speech reviewreviewer pool

Serving

Inference on our own hardware, behind our own gateway, with the data staying here.

inference gatewaymodel registryon prem serving

Observability

Every node, job and run scraped and dashboarded, so nothing about a run is a guess.

PrometheusGrafanapushgatewayGitHub
6 + frontierGPU nodes in house, with frontier GPU on call
22 TBusable on premise storage for corpora and checkpoints
5 smetrics scrape interval across every node and job
21repositories: models, tooling, consoles, infrastructure

We are building the capability, not just the product.

If that is the kind of work you want to do, or the kind of partner you have been looking for, we would like to hear from you.

hello@kapllan.ai →