Materials & process development
Composition, synthesis, characterization, and scale-up—from 2 L to 50 L.
AI for Science
G1Brain
A Data-Driven AI Flywheel
Scientific knowledge. Preserved with context. Readable by AI.
G1Brain is Group1’s internally developed scientific knowledge platform. Built through four years of potassium-ion development, it makes experimental evidence, provenance, and scientific reasoning AI-readable.
Group1’s potassium-ion science traces its origins to the pioneering battery research of John B. Goodenough, recipient of the 2019 Nobel Prize in Chemistry.
Every experiment adds evidence. Every decision builds on it.
4 years
Research coverage, 2022 to Q4 2026
55,000+
Research files preserving the experimental record
8,618
Active structured records, including templates
600+
Technical reports preserving scientific reasoning and semantic context
Developed internally. Used in practice.
Built to accelerate our own science.
We built G1Brain to solve our own development challenges. It helped us rapidly commercialize potassium-ion materials, electrolytes, and batteries.
Reports and scientific discussions retain semantic context: what a result meant to the researchers, which explanations they considered, and what they chose to investigate next.
Cell performance
Every test adds to the record.
Cell testing produces the measurements behind performance comparisons and further investigation.
Materials and process
Physical science. Digital knowledge.
Pilot-scale synthesis connects materials development with the process conditions and observations that give data context.
The same chemistry at 2 L and 50 L.
Predictive models often break down when a process changes scale. The record holds synthesis runs at both, with conditions and outcomes.
What failed, and why.
Failed experiments and underperforming cells, kept with the diagnosis and the corrective action.
Similar conductivity. Different results.
Three isomers of one electrolyte molecule measured alike, then performed differently in cells. The record holds the measurements, the cell data, and the reasoning.
Four layers of scientific knowledge
The scientific record—and the reasoning behind it.
01 / Original evidence
Instrument files, spectra, images, and cell-test data.
02 / Structured records
Formulations, process conditions, and results.
03 / Scientific interpretation
Reports and discussion that hold the reasoning.
01
Raw scientific data
Measurements, instrument files, and images preserved from the original experiments.
55,000+
research files
02
Labeled and structured data
Records connect results to their materials, conditions, and experimental history, with provenance preserved.
17.29M
Measurement references in structured records, not verified measurement values
03
Human scientific interpretation
Scientist-led reports preserve results, interpretations, and the reasoning behind next steps. Recent reports include AI-assisted preparation.
600+
Technical reports preserving scientific reasoning and semantic context
Grounded in physical science
From materials to cell performance.
Materials and process
Synthesis recipes, processing conditions, thermal analysis, diffraction, microscopy, and process development.
Electrolyte design
Formulation choices, physical properties, spectroscopy, and comparisons of cell behavior.
Cell engineering and testing
Electrode preparation, assembly, cycling, degradation, and failure analysis across coin, pouch, and cylindrical formats.
The platform in practice
From evidence to the next experiment.
Researchers followed one batch of battery material from thermal characterization to cell testing. Each step is a record, and each record links to the next.
01
Characterize the material
Thermal measurements on one synthesized batch
02
Build the electrode
Slurry and electrode records, with their conditions
03
Test the cells
Cycling results, tied back to the batch
04
Decide what comes next
The team's discussion of what the result meant
G1Brain / One connected thread
One material batch connects the original measurements, build records, and the team’s interpretation.
A second traced example follows an electrolyte design study from formulation, through physical property measurements, to cell cycling. Not every experiment is linked this completely. The linked threads are where the record is richest.
Scientific knowledge made AI-readable
Make each development cycle build on the last.
AI for Science needs more than published results. It needs the experimental record: what was tried, what failed, and why researchers chose the next test. G1Brain holds that record for potassium-ion batteries, from materials synthesis to cell performance.
Experiment-planning agents
Ask what to test next. Compare the plan with what the scientists chose.
Evidence-grounded answers
Answers that cite the original measurement.
Synthesis and scale-up guidance
Recipes and conditions, from bench to pilot scale.
Formulation-to-performance models
From electrolyte formulation to cell behavior.
Learning from failed approaches
Negative results the literature rarely reports.
Reading instrument data
Spectra and cycling curves, paired with interpretation.
G1Brain preserves the evidence and reasoning that researchers and AI need to carry knowledge forward.
Recover knowledge
Find the evidence behind an answer.
Recover original results alongside their source records and scientific interpretation.
Prediction across scale
Relate synthesis recipes and process conditions to material quality as batches move from 2 L to 50 L.
Learning from failed approaches
Use unsuccessful experiments and failure analysis, which published literature rarely reports, to show models where the boundaries lie.
Reading instrument data
Pair spectra, diffraction patterns, micrographs, and cycling curves with the scientists' written interpretation.
Connect evidence
Follow the relationships.
Trace how materials, conditions, and decisions relate to experimental outcomes.
Evidence-grounded answers
Answer a technical question by recovering the reported result, citing the original measurement, and explaining the recorded interpretation.
AI-automated laboratories
Ground an automated lab's planning in a real record of conditions, outcomes, and the decisions that followed.
Guide development
Build on what is already known.
Use prior successes, failures, and interpretations to inform the next experiment.
Beyond batteries
Potassium-ion technology is the first application. The underlying approach—preserving experimental evidence, scientific reasoning, and decisions in an AI-readable record—is designed to support scientific development across disciplines, with structure and validation adapted to each field.
In use today. A foundation for what comes next.
Group1’s research agents already query the record and scientific literature. Model training and laboratory automation are future applications.
Source records are read-only.
Quantitative claims carry sample count and spread.
Comparisons across records are computed, not estimated.
Answers cite the records they draw on.
Connections vary by experiment.
Built into the work
Four years in the making.
Built through four years of use, informed by our team’s experience building scientific research organizations.
2022
The record starts
Experiments, technical reports, and research discussion are kept from the first year of work.
2023
Structured records begin
Formulations, process conditions, and results go into a lab informatics platform as the work happens.
2024
The record expands
The scientific record extends from coin cells to Group1’s first 18650 cylindrical cells.
2026
Agents work on the record
AI research agents query the structured records and the literature alongside the team.
Built at Group1. Designed for broader scientific use.
A platform for scientific development.
Potassium-ion technology is our first application. The same approach—preserving evidence, provenance, and interpretation for AI—can support development across scientific disciplines.
Scientific overview
Three pages on the record, its scale, connected evidence, and example applications. PDF, Q4 2026.
Download the overview (PDF)Connect with Group1
Discuss the platform, its use at Group1, and its broader scientific potential.
Contact Group1Download Scientific Overview (PDF) ↓
Contact Group1 →Three pages on G1Brain’s scientific record, scale, connected evidence, and potential AI applications. Q4 2026.