Scientific task
Build scientific AI
with the whole lab record.
Experimental data, scientific expertise and evaluations grounded in real lab work. Start with the questions your model needs to answer, then define the measurements, context and review the work requires.
Our approach
Start with the
scientific question.
Useful data begins with the task. Connect what your model needs to learn with the evidence, context and permissions behind each record.
Experimental context
Keep methods, samples and controls connected to each result.
Permitted use
Agree the scope and rights before preparation begins.
Our offerings
How we partner with AI teams.
Start with existing data, a scientific workflow or a question worth testing.
01 Data · Experimental records
Measurements with
their scientific context.
Unpublished assay results, instrument data and experimental records across biology, chemistry and materials. Positive, negative and inconclusive outcomes stay linked to the methods behind them.
- Native files and measured results
- Sample identity and preparation
- Methods, controls, repeats and QC
- Agreed training or evaluation rights
From measurement to meaning
Keep the record connected.
01
Sample
Identity · preparation · provenance
02
Method
Protocol · conditions · instrument
03
Measurement
Raw files · results · units
04
Interpretation
Controls · repeats · quality checks
02 Evals · Scientific workflows
Explore what good
scientific work looks like.
Bring a workflow, a model behaviour or a question you want to investigate. Together, we can explore relevant tasks, scientific review and ways to assess the results.
- The scientific question and task
- Evidence available to the model
- Criteria for reviewing its reasoning
- Uncertainty and limitations to examine
Questions to explore
What supports the answer?
Evidence
Does the answer trace back to the experimental record?
Reasoning
Does the interpretation account for methods and controls?
Limits
Does it distinguish what is known from what remains uncertain?
03 Expertise · Scientific judgement
Bring domain judgement
into the work.
Explore expert review of experimental records, scientific reasoning and model outputs. Scope the field, review criteria and specialist experience with us.
- Relevant scientific methods and experience
- Record interpretation and annotation
- Review of model-generated reasoning
- Explanations grounded in the evidence
A traceable line of reasoning
From record to rationale.
Source record
Measurements and context
Scientific review
Methods and interpretation
Written rationale
Evidence and limitations
04 Experiments · New measurements
New measurements,
designed around the gap.
When an existing archive cannot answer the question, discuss a targeted experimental programme. Methods, laboratory capability, scope and acceptance criteria are agreed before work begins.
- A defined scientific question
- Methods, samples and controls
- Required measurements and repeat design
- Feasibility, rights, timing and acceptance
Define before generating
Design around the question.
Test conditionsControlsRepeats
MethodReadoutQuality criteria
Soon
Scientific environments
We are exploring replay tasks and simulated lab workflows for AI agents, with scientific constraints and traceable outcomes.
Keep the scope clear.
Simulation fidelity, validation and access to physical experiments need separate review. Share the workflow you would like to explore.
Explore an ideaStart with clear rights
and defined access.
Agree the records, recipients and permitted uses before release.
Scoped records
Define which studies, samples and files belong in the engagement.
Scientific traceability
Keep methods, controls and quality flags linked to the measurements.
Permitted uses
Agree recipients, training or evaluation uses and any exclusivity.
Release approval
Confirm the dataset and transfer conditions before sharing it.
Where scientific AI can help
Data and expert review for model training and evaluation across scientific fields.
Medicinal chemistry
- Structure–activity relationships
- Biochemical and cellular assays
- Compound properties
- Selectivity and off-target activity
Pharmacology & DMPK
- Dose–response analysis
- ADME and pharmacokinetics
- Exposure and response
- Study design review
Cell biology & cytometry
- Cell-based assays
- Microscopy and image analysis
- Flow cytometry
- Cell phenotype annotation
Analytical chemistry
- Chromatography
- Mass spectrometry and spectroscopy
- Method validation
- Peak identification and integration
Protein science
- Protein expression and purification
- Biophysical characterisation
- Binding and activity assays
- Stability and formulation
Computational biology
- Genomics and transcriptomics
- Bioinformatics workflows
- Experimental design and statistics
- Sequence and sample quality control
Materials science
- Materials characterisation
- Property and performance testing
- Processing conditions
- Batch variation and failure analysis
What does your model need next?
Tell us the scientific task and the gaps you want to close.
Get in touchFrequently asked questions
Your scientific AI
questions, answered.
Where do we start?
With a scientific task, a data specification or a workflow you want to investigate. We will discuss scope, available records and the next useful review.
Can we discuss evaluations before defining a benchmark?
Yes. An initial conversation can explore the scientific question, the relevant workflow and possible assessment approaches.
Can you source both existing and newly generated data?
We discuss licensing existing laboratory records and commissioning new experiments. Feasibility, rights, cost and timing depend on the project.
How are rights and access handled?
Permitted uses, recipients and release conditions are agreed for each engagement.