Pipelines as decisions
A pipeline is not a script — it's a choice of modality, granularity, fusion and backend. Brain-AI makes that choice explicit, serializable and searchable.
SolarpunkWorksPvt Ltd
Decision intelligence for ML
Brain-AI
An open-source engine that decides how multimodal machine-learning pipelines should be built — then executes those decisions.
MultimodalModularOpen source
The idea
Not another model — a way to choose.
Most ML frameworks train models. Brain-AI asks a different question: given your data and goal, which pipeline should you build? It separates the decision from the execution — exploring which modality, granularity and fusion combinations actually work, then running them through modular components.
Built for ML engineers and researchers who want to systematically compare pipeline configurations instead of guessing.
What it does
Decide, execute, compare.
Explore the whole space
Generate every valid combination of strategies and backends, then run them side by side — instead of betting on one guess.
Granularity, aligned
Resample, pooling or attention — align multimodal inputs the way your data actually behaves.
Fuse on your terms
Early, late or intermediate fusion, with your overrides always winning. Built with scikit-learn; ready for AutoGluon and TPOT.
Every run, reproducible
Batch experiments, a leaderboard and MLflow tracking — each run produces metrics and a DAG you can defend.
Built for LLM agents
Structured, safe entry points let an AI generate specs, run experiments and compare models — Brain-AI is designed to be orchestrated, not just used.
What's next
Ablation studies, designed by AI.
Which component of a pipeline actually matters — the granularity strategy, the fusion point, the backend? Our upcoming research uses AI to design and run ablation studies systematically: isolating one decision at a time, measuring its effect, and building evidence about why a pipeline works, not just that it works.
Brain-AI's decision-object design makes every pipeline a controlled experiment waiting to happen.
Follow the research
Get it
On PyPI today.
Apps for every platform, soon.
Brain-AI is free and open source, published as brain-automl — pip-installable today, with the full decide-align-fuse-evaluate loop, MLflow tracking and DAG artifacts for every run.
A native app for macOS, Windows, Linux and the web is on the way — the same decision engine, one download, no Python required.
Start in one line
pip install brain-automlFree & open source · Python ≥ 3.9 · scikit-learn included · AutoGluon / TPOT optional · MLflow built inComing soon: macOS · Windows · Linux · Web