Independent AI & software lab

AI research,
built into
working systems.

Itinn is an independent AI and software lab developing its own products. We combine model adaptation, controlled automation and software engineering.

See the work

Built around reliability, cost and privacy.

Form study / 001 Interactive

Change the form

A small experiment in geometry. Try the slider.

01 / Selected work

From the workbench.

How we evaluate models
and control the actions they propose.

Model evaluationClient work

Finding the right model for the job.

We’ve compared AI models and reasoning effort settings for client workflows, looking for the best value at the quality each task needs.

How we approach it

We start with the client’s tasks and define what an acceptable result looks like. We compare model and effort combinations against those requirements, then examine the cost of getting a useful result.

The findings help decide where extra reasoning is worth the cost and where a lighter setup is enough. Client data and detailed results stay private.

Public exampleSynthetic data

An action policy, made visible.

An agent’s proposed action needs to pass explicit checks before it can run. This small example shows how request validation and granted permissions affect that decision.

Try the policy checks

A synthetic example of checking an agent’s proposed action before it can run.

Proposed action

{
  "action": "read_summary",
  "project": "demo"
}

Granted access: read_summary on demo.

Valid request: Passed
Within granted access: Passed

Allowed by this policy

Access rule: reading the demo summary is within the granted permission.

Illustrative policy evaluated in your browser. Synthetic data; no action is executed.

In a deployed system, these checks belong where the action is executed.

02 / Research & engineering

Built around the job.

We choose the approach around the quality a task needs,
the cost it can support and the data it must protect.

01Area of work

AI agents in practice

We’ve built agent workflows with scoped access, controlled tool use and human oversight. The focus is useful actions within clear limits.

Scoped access · Tool use · Human oversight

02Area of work

Engineering with AI

Which model fits the job? We compare approaches on defined tasks, looking at correctness, time and cost.

Model evaluation · Development workflows

03Area of work

Product engineering

We turn technical ideas into working software. Prototypes help us test the workflow, integration and practical value before committing to a product.

Prototypes · Interfaces · Live systems

04Area of work

Private & local models

We’ve adapted open-weight models to specific tasks with LoRA and evaluated them for local and private deployment. Our work also includes statistical machine learning and deterministic guardrails.

Model adaptation · Task evaluation · Private deployment

Methods we’ve applied

From general models to specialist systems.

Our experience spans explicit rules, compact predictive models and adapted language models.

Deterministic guardrails: Explicit rules, schema validation and permission checks put defined boundaries around model inputs, outputs and actions.
Fast statistical machine learning: Compact models for classification, scoring and routing, selected against accuracy, latency and compute requirements.
Generative LLM fine-tuning with LoRA: We adapt open-weight models to a task’s behaviour, terminology and output requirements, then evaluate them on examples kept separate from training.

We compare quality, response time and running cost. Local and private deployments are assessed against each project’s privacy and data-handling requirements.

03 / The lab

Research with
a product in mind.

We’re building Itinn around our own software products, with applied research at the centre of how we develop them.

Our experience with cloud systems, mobile applications, payments and access control shapes the problems we choose and the way we build.

We look for recurring problems with a clear cost to the people doing the work. We test technical feasibility through prototypes. Further development depends on evidence of practical value, feedback from intended users and a credible path to a business.

We keep a record of what fails as well as what works. That evidence guides where we invest our effort next.

04 / Get in touch

Let’s talk about
what comes next.

For investment conversations, research partnerships and technical collaboration, get in touch.

We can share more about the lab’s background and direction in a private conversation.

itinn@itinn.eu

AI research,built intoworking systems.