Predict material properties from the formulation, process, and test data your team already has. A few hundred rows is enough, and every upload leads to the reasoning and the next experiment.
Set the formulation, run the test, and when it falls short, adjust by intuition and go again. Nobody knows in advance how many rounds that takes.
Trial and error → time and cost pile up
The formulation is adjusted by intuition, round after round, until the target is met
We predict bulk-scale properties from a small amount of experimental data. Not large-scale compute searching for novel materials — answers about the formulations a company already owns.
Select a node to see each stage
First it judges whether your data can predict the property at all. When the answer is no, we say so up front.
TendiHub is a materials R&D platform that predicts properties from the experiment data a company already has. Centi, its first product, is a prediction SaaS materials teams run in the browser with no dedicated AI staff.
| ID | Resin | Filler | Temp | Density |
|---|---|---|---|---|
| PP-01 | 32.5 | 18.0 | 230 | 1.04 |
| PA6-04 | 28.0 | 24.5 | 255 | 1.18 |
| PP-07 | 35.0 | 15.0 | 225 | 0.98 |
| PA6-09 | 26.5 | 28.0 | 260 | 1.24 |
Showing 4 / 400 rows · latest first
Keep your format. We start by organizing scattered records.
The composite validation used 400 rows with 100+ features.
| ID | Resin | Filler | Temp | Density |
|---|---|---|---|---|
| PP-01 | 32.5 | 18.0 | 230 | 1.04 |
| PA6-04 | 28.0 | 24.5 | 255 | 1.18 |
| PP-07 | 35.0 | 15.0 | 225 | 0.98 |
| PA6-09 | 26.5 | 28.0 | 260 | 1.24 |
Showing 4 / 400 rows · latest first
| ID | Resin | Filler | Temp | Density |
|---|---|---|---|---|
| PP-01 | 32.5 | 18.0 | 230 | 1.04 |
| PA6-04 | 28.0 | 24.5 | 255 | 1.18 |
| PP-07 | 35.0 | 15.0 | 225 | 0.98 |
| PA6-09 | 26.5 | 28.0 | 260 | 1.24 |
Showing 4 / 400 rows · latest first
KoMaP 2025 Grand Prize, 1st of 61 teams. 8 composite properties at R² 0.93+.
The validation predicted 8 properties simultaneously.
| ID | Resin | Filler | Temp | Density |
|---|---|---|---|---|
| PP-01 | 32.5 | 18.0 | 230 | 1.04 |
| PA6-04 | 28.0 | 24.5 | 255 | 1.18 |
| PP-07 | 35.0 | 15.0 | 225 | 0.98 |
| PA6-09 | 26.5 | 28.0 | 260 | 1.24 |
| N-01 | 33.0 | 16.5 | 232 | — |
| N-02 | 27.5 | 26.0 | 258 | — |
Showing 4 / 400 rows · latest first
| ID | Resin | Filler | Temp | Density |
|---|---|---|---|---|
| PP-01 | 32.5 | 18.0 | 230 | 1.04 |
| PA6-04 | 28.0 | 24.5 | 255 | 1.18 |
| PP-07 | 35.0 | 15.0 | 225 | 0.98 |
| PA6-09 | 26.5 | 28.0 | 260 | 1.24 |
| N-01 | 33.0 | 16.5 | 232 | — |
| N-02 | 27.5 | 26.0 | 258 | — |
Showing 6 / 400 rows · latest first
Test only the suggested candidates. Screening runs shrink into predictions.
Experiment results accumulate as training data.
2025 KoMaP AI Competition Grand Prize, Minister of Trade, Industry and Energy Award — team AIMSE (now the Tendilab founding team), 1st of 61 teams Read the article ↗
All 8 at R² 0.93+
| Metric | TendiHub model |
|---|---|
| Tensile strength | 0.942 |
| Tensile modulus | 0.955 |
| Flexural strength | 0.938 |
| Flexural modulus | 0.931 |
| ILSS | 0.939 |
| Impact strength | 0.947 |
| Hardness | 0.958 |
| Density | 0.943 |
test R². The 8 composite properties were validated by a government research institute. Bar charts do not start at zero.
Give it a keyword and it reads open-access papers into a single table of composition, process, and properties — training data for the foundation model.
Enter a keyword → formulation data extracted from open-access papers
| Composition | Processing | Property |
|---|---|---|
| Epoxy / CF 60 wt% | 180 °C, 2 h | 1.94 GPa |
| Epoxy / CF 55 wt% | 160 °C, 3 h | 1.71 GPa |
| Epoxy / CF 50 wt% | 180 °C, 1 h | 1.52 GPa |
Values shown in the table are format examples.
MELD: Multimodal Extraction of Machine-Learning-Ready Materials Data from Scientific Literature ↗Submitted to EMNLP 2026
Properties a formulation table cannot explain are learned together with data at smaller scales.
Tabular composition and process data. The scale we have validated today.
Formulation sheets, process variables, and the properties that matter all differ by industry. We take your format as it is and fit the model to your domain.
Polymer, additive and filler ratios; extrusion temperature; thermoforming and injection molding conditions
Tensile strength and modulus, flexural strength and modulus, ILSS, impact strength, hardness, density
It starts with a diagnosis. Scale comes after something is proven to work.
Leave your email and a line about the material or problem you work on. We will review whether it is a predictable problem and reply.