Generative design for cultivated bio-materials
Celleste is not a general materials database. It is built for the composition variables and property targets central to bio-material development.
Generate candidates from your variables, not ours
Set ingredient variables and ranges for your material system. Celleste creates candidates within your lab constraints, favoring compositions likely to meet target properties. Export the ranked results to your lab notebook.
- Handles up to 12 composition variables per material system
- Scores candidates by predicted property alignment
- Exports CSV, JSON, and common lab notebook formats
Predict properties before running a batch
For every candidate formulation, Celleste provides predicted values and confidence intervals for key properties: tensile strength, elongation, biodegradation rate, and cell adhesion score. Confidence widens for novel compositions and narrows as your lab gathers measurements.
Fits your lab process
Morning: set the design problem
Set the target property profile and composition limits for the week's formulation focus.
Midday: review Celleste candidates
Review the ranked candidate list. Filter by confidence threshold or a property constraint, then flag formulations worth running.
End of week: close the loop
Once batches run, record the actual results. Celleste updates the model for that material system and improves future candidates.
Celleste uses data your lab already has
You can begin without a large historical dataset. Celleste starts with literature priors for your material class and improves with real batch data. Minimum viable input: five or more past batches, composition records, and at least one measured property per batch.
What you bring
- Past batch records: composition, measured properties
- Target property profile for the new formulation
- Composition variable ranges and constraints
Celleste returns
- Ranked formulation candidates with predicted properties
- Confidence intervals for each property and candidate
- Candidate comparison against your target profile