Browser storage is unavailable. Keep this page open and export your collection before leaving.
We're scaling self-improving AI scientist teams to ideate/experiment/discover, generating new knowledge autonomously. Their discoveries reduced ARC-1/2 costs by 10,000× while maintaining state-of-the-art accuracy and yielded material advances on combinatorial ML problems. We believe that AI is bounded not by compute, but ideas. Scaling to millio
We're scaling self-improving AI scientist teams to ideate/experiment/discover, generating new knowledge autonomously. Their discoveries reduced ARC-1/2 costs by 10,000× while maintaining state-of-the-art accuracy and yielded material advances on combinatorial ML problems.
We believe that AI is bounded not by compute, but ideas. Scaling to millions of sessions of discovery enables leveraging its own discoveries and presents a novel training source capable of ushering a new era of self-reliant foundational models.
Previously rekursiv
Ex-Google, Ex-Deepmind, Ex-LumaAI staff research scientist for a combined total of 14 years. Led foundational model pre-training at Luma AI. Creator of TensorFlow Probability (1M+ downloads / month), built Veo's first prototype, contributor to Gemini and VideoPoet (ICML 2024 Best Paper). 11,600+ citations. Now co-founder of rekursiv.ai, automating the science of ML research.
Ex-Google, Ex-LumaAI scientist for a combined total of 7 years. First author of VideoPoet (ICML 2024 Best Paper), led the World Models team at Luma, and built large-scale multimodal LLMs and diffusion models for image, video, audio, and text. Now co-founder of rekursiv.ai, automating the science of ML research.
Founded 2026 · Source
Source-based assessment · automated evidence rules
The latest website check does not confirm this company’s website.
Maturity is an evidence-based assessment, not a success rating. It may understate the current stage where evidence is limited. Coverage records documented areas, not a complete or independently audited company history.
Agents that curate your data, post-train open-weight models, evaluate and serve them. You own the weights.
Product
Same product type · Infrastructure
Aviro designs challenging worlds for AI models to explore and solve problems, delivered as post-training datasets and reinforcement learning environments.
Product
Same product type · Infrastructure
Make models like it's nothing. The custom model factory for cheaper, better, faster AI.
Same product type · Infrastructure
Ship AI agents that compound. Full-fidelity traces, plain-English alerts, auto-authored skills, and regression-proof evals — OpenTelemetry-native.
Founders · Product
Same product type · Infrastructure
Custom conversational audio datasets, expert annotation, and licensed voice actor data for training and evaluating speech and voice AI models.
Founders · Product · How it works
Same product type · Infrastructure
BAML is one toolchain from the compiler to production. Build, test, trace, query, and alert on your code with tools that all read the same types.
Founders · Product
Same product type · Infrastructure
