Autoscience Institute has recently secured a funding round, backed by Perplexity to advance its automated AI research lab. Autoscience Institute builds ML models that help to accelerate machine learning research. Over 2000 research papers are published on arXiv every week. Autoscience’s Mira model reduces weeks of paper review research into a single data driven workflow. It also allows researchers to conduct cross-codebase research, identifying which methods reliably produce real world implementations. Autoscience’s ML model Carl, reads research papers and identifies hypothesis that would advance ML research. It then conducts experiments to test the hypothesis and eventually, synthesizing its findings and organizes its experiment results into a full length research paper.

The goal is to compress years of research into a matter of months. Their managed service deploys hundreds of “automated AI research Scientists” to continuously improve and ship updates to customer models simultaneously.

Their system has already achieved 2 remarkable milestones. It produced the first peer reviewed scientific paper authored by an autonomous lab. It also earned a silver medal in a Kaggle machine learning competition (Kaggle Santa 2025) against 3300 human led teams. It is the first AI model to win a Kaggle competition.

Moving forward, Autoscience is currently expanding its engineering team and seeking partnerships with large enterprises that want to keep pace with the rapid evolution of model performance without relying solely on understaffed human ML teams.