# Tacit Intelligence

> Capture the expert judgment that cannot be written down, taught in training, or handed over in two weeks, then deploy it across your whole team.

The Tacit Intelligence Company is a New Zealand company that trains specialised language models on the judgment of an organisation's best people. Contact: hello@tacitintelligence.co.

## Pages

- [Home](https://www.tacitintelligence.co): What Tacit Intelligence does, how an engagement works, the domains served, and how to get started.
- [Case Studies](https://www.tacitintelligence.co/case-studies): Deployments of specialised language models, with measured results.
- [Nourish case study](https://www.tacitintelligence.co/case-studies/nourish): Scaling high-quality nutrition consultations with a specialised language model and human-in-the-loop review.
- [Otago Polytechnic case study](https://www.tacitintelligence.co/case-studies/otago-polytechnic): Research partnership using simulated clients to teach coaching judgment, measured turn by turn.
- [Field Notes](https://www.tacitintelligence.co/field-notes): Research notes and writing from the team.
- [Frontier Labs](https://www.tacitintelligence.co/frontier-labs): Reinforcement learning environments and expert-calibrated ground truth for frontier laboratories.
- [About](https://www.tacitintelligence.co/about): The team, the story, and the values behind the company. Based in New Zealand.
- [Methodology](https://www.tacitintelligence.co/methodology): The training-methodology spectrum, from prompting to reinforcement learning with verifiable rewards.
- [Developers](https://www.tacitintelligence.co/developers): Developer hub linking methodology, base models, and frontier-labs resources.
- [Base Models](https://www.tacitintelligence.co/base-models): The open-weight base model families we customise into specialised models, and the platforms we deploy to.

## Field notes

- [The 8B SLM That Beats Frontier AI on HealthBench](https://www.tacitintelligence.co/field-notes/trained-specialised-language-model-healthbench): We trained an 8B parameter SLM, specialised language model, and evaluated it on the hardest physician-validated subset of HealthBench. It scored 96.5% accuracy and retained 72.4% under a worst-of-16 stress test.
- [Tacit and AI Sovereignty: Owning the Translation Layer](https://www.tacitintelligence.co/field-notes/tacit-and-ai-sovereignty): Defining AI sovereignty as owning the entire AI stack makes it sound impossible for New Zealand. This is the wrong definition. Sovereignty lives in the translation layer between foundation models and trusted real-world deployment.
- [What Is Tacit Knowledge?](https://www.tacitintelligence.co/field-notes/what-is-tacit-knowledge): Your best people make calls the manual can't explain. That judgment has a name - tacit knowledge - and it was learned through reps, not reading. It is the most valuable asset on your payroll, it walks out the door when people do, and it is the one training corpus your competitors cannot buy.

## Optional

- [Privacy Policy](https://www.tacitintelligence.co/privacy): How we handle personal information under the New Zealand Privacy Act 2020.
- [Terms of Service](https://www.tacitintelligence.co/terms): Terms governing use of the platform and website.