AutoScientist by Adaption Labs

AutoScientist is a system from Adaption Labs that automates the full research loop behind training machine learning models. Instead of manually tweaking datasets and hyperparameters, you tell it what outcome you want — like improving accuracy on a specific task — and AutoScientist handles the iterative experimentation. It co-optimizes both your training data and the training recipe (the sequence of steps, data mixes, and hyperparameters) until the model converges on your goal. The result is a deployed model that behaves the way you intended, without requiring deep expertise in model training.

What sets AutoScientist apart is that it treats data and recipe optimization as a single, coupled problem rather than separate steps. Most workflows require you to curate data first, then train, then adjust — AutoScientist loops through both simultaneously. The page reports significant relative improvements across domains: 78% overall, with finance seeing 130%, technology 112%, enterprise 89%, and medical 84%. It's designed to work across verticals without needing retraining for each domain, so you can use the same system for a medical imaging model and a financial forecasting model.

This tool is best for teams that need to train custom models but lack the time or expertise to manually run hundreds of experiments. It's useful for researchers, ML engineers, and product teams in regulated industries like healthcare and finance where model behavior matters. If you're building a specialized model and want to skip the trial-and-error phase, AutoScientist aims to do the science for you.