Formulation

AI Formulation Tool

The in-house ML lab. Bayesian optimization over the test database proposes the next experiment, ranked by expected information gain and the uncertainty of its prediction.

Training samples

842

Tests feeding the model

Model freshness

12 min

Last re-train

Candidates ranked

Awaiting review

Avg predicted lift

+7.4%

vs. current best recipe

Optimization target

Editable · applies to all candidates

Compression ≥ 60 MPa

Water ≤ 3.5%

UV ≥ 8.5

Ranked candidates

Sorted by expected information gain

1CAND-042Compression (Facade)

DF 55% / lignin 18% / PF binder +3%

Feedstock · Douglas Fir + Organosolv lignin

Predicted

61.8 MPa

[58.2, 65.1] 95% CI

Expected information gain84%

2CAND-041Water absorption (Facade)

DF 40% / CNF 6% / silane-treated

Feedstock · Douglas Fir + TEMPO-CNF

Predicted

2.4 %

[2, 2.9] 95% CI

Expected information gain78%

3CAND-040Thermal conductivity (Insulative)

Cedar 70% / PU binder / low-density

Feedstock · Western Cedar + PU

Predicted

0.034 W/m·K

[0.031, 0.037] 95% CI

Expected information gain71%

4CAND-039STC rating (Acoustic)

Spruce 50% / CNC 4% / porous mix

Feedstock · Spruce + CNC

Predicted

44 dB

[41, 46] 95% CI

Expected information gain62%

5CAND-038UV stability (Facade)

Oak 45% / kraft lignin 25% / pigment+1%

Feedstock · Oak + Kraft lignin

Predicted

9.5 /10

[9.1, 9.8] 95% CI

Expected information gain55%