We operate at the convergence of autonomous robotics, physical chemistry, and machine learning — pursuing fundamental science through unprecedented experimental throughput.
Our electrochemistry program centers on highly parallelized, automated potentiostatic and galvanostatic platforms capable of running hundreds of electrochemical experiments simultaneously without human intervention.
We employ multichannel bipotentiostats integrated with robotic liquid handling, inline spectroelectrochemistry, and impedance spectroscopy modules to construct comprehensive datasets of electrode-electrolyte interfacial behavior at unprecedented scale.
Key application domains include electrocatalysis for energy conversion, electrochemical CO₂ reduction, nitrogen fixation, and synthesis of high-value organic compounds under controlled potential conditions.
The BRAINLAB catalysis program develops robotic workflows for high-throughput synthesis and characterization of heterogeneous, homogeneous, and single-atom catalysts. Our combinatorial screening infrastructure enables exploration of multi-component composition spaces at rates impossible by conventional methods.
We focus on electrocatalysts for the hydrogen evolution reaction (HER), oxygen evolution reaction (OER), CO₂ reduction, and nitrogen fixation — guided by density functional theory pre-screening and machine learning surrogate models trained on our proprietary experimental datasets.
True autonomous experimentation requires more than robotic hardware — it demands intelligent decision architectures that perceive experimental outcomes, update probabilistic models, and propose the next experiment without human oversight.
Our Bayesian optimization framework integrates Gaussian process regressors, multi-objective acquisition functions, and constraint-handling mechanisms that navigate complex synthesis parameter spaces. We have demonstrated fully autonomous campaigns of up to 10,000 consecutive experiments on our ATLAS platform.
We also pioneer multi-fidelity experiment design, combining cheap computational predictions with expensive physical experiments in a unified active learning framework that minimizes total resource consumption.
BRAINLAB's AI program develops bespoke machine learning architectures for the unique demands of chemical data: sparse, multi-modal, physically constrained, and generated by an autonomous system that it is simultaneously trying to guide.
We develop graph neural network potentials, equivariant message-passing networks for molecular property prediction, and large-scale foundation models pre-trained on our proprietary electrochemical datasets containing over 2.1 million experimental data points.
Our data pipeline handles real-time ingestion from 24 simultaneous analytical channels, automated quality control, and seamless integration with active learning loops that feed new experimental requests back to the robotic platform.
Continuous real-time monitoring across voltage, current, pH, and temperature channels — streamed and analyzed autonomously at 10 kHz sampling rates.