KROG Lab
Domain experts train KROG-AI
KROG Lab is where senior lawyers fine-tune open-weight language models on the KROG rulebook. Not prompted — trained. Per jurisdiction, signed by named experts.
Why training, not prompting
The KROG rules are a new, narrow rulebook. General models guess. A fine-tuned model knows — because it has actually seen the rules, examples and counter-examples during training.
How we do it
We use Tinker from Thinking Machines to train open weights (Llama, Qwen) on curated datasets. One base model + LoRA adapters per jurisdiction. Every training run and evaluation is documented.
Who trains the models
KROG Fellows are senior lawyers in the network who take responsibility for a model version's quality in a given country. They contribute Q&A pairs, rule annotations and counter-examples, and sign off on each release.
KROG Fellows
Domain experts who train and sign off on the models.
We are appointing the first Fellows. Reach out if you want to be part of the first cohort.
Models
Each published model has a model card with base weights, jurisdiction, trainers and an evaluation summary.
No published models yet. The first release is on its way.
Become a KROG Fellow
Are you senior in one of the KROG practice areas? We are admitting a limited number of Fellows for the first round of model training.
Apply for Fellowship