Rootcomputer

Rootcomputer is an independent AI research lab focused on building small language models and the custom training pipelines, datasets, and alignment strategies required to build them from scratch.

Our work is experimental and open-research focused. We study how architectural decisions, data composition, and post-training alignment shape behavioral stability, reasoning, and cognition at small scale.


What You’ll Find On This Channel:

 • Model Architecture & Deep Dives: Technical breakdowns and evaluations of our open-weight models, including Haiku, Epos, and Tanka.

 • AI Research & Experiments: Explorations into compact transformer design, context length extension, failure analysis, and human-AI coexistence.


Connect & Explore:

 • Website: https://rootcomputer.dev

 • Hugging Face: https://huggingface.co/rootcomputer

 • Research & Benchmarks: https://rootcomputer.dev/research


Building AI small on purpose.