Teary homepage
Scores
Visualisation app
Product example

Teary

A disease classifier and a keepsake, built for your tears.

Inspiration

A tear isn't just saltwater, it's a biological fingerprint. When tears dry, they form dendritic crystal patterns shaped by your biology, mood, and the moment itself. We saw two possibilities in one drop: a diagnostic signal for disease, and an artifact of memory. Teary is both.

What it does

Teary turns a single teardrop into two things:

A disease classifier — upload an image of a dried tear and our model predicts one of five classes: Healthy, Diabetes mellitus, Glaucoma, Multiple sclerosis, or Dry eye syndrome.
A keepsake — the same dendritic pattern gets imprinted onto everyday objects (an etched optical crystal plaquette, a glazed ceramic plate, a laser-etched notebook cover) so tears of joy become something you can hold forever.

How we built it

Fine-tuned EfficientNet-V2 on dendritic tear-crystal images across five disease classes.
Designed a home collection kit — since UPJŠ scientists confirmed tears are biologically stable in transit and resistant to heat, users can cry at home, mail the sample, and get results plus their keepsake back.
Prototyped the keepsake line: optical crystal plaquette, dendrite-under-glaze ceramic plate, and laser-etched notebook cover.

Challenges we ran into

Small, imbalanced dataset — tear-crystal imagery is niche, so we had to be careful with augmentation and validation splits to avoid overfitting.
Five-class separation — some disease signatures look visually similar under dendritic formation, which pushed our macro-F1 down on the full five-class task.
Bridging biology and design — making a medical tool that also feels like something you'd want on your shelf required constant back-and-forth between ML, UX, and product.

Accomplishments that we're proud of

94.1% macro-F1 on binary classification (healthy vs. diseased)
82.4% macro-F1 on three-class
64.7% macro-F1 on the full five-class problem

What we learned

Tears carry far more structured information than we expected, dendrite morphology really does track with systemic disease.
Sample stability matters: a diagnostic is only useful if people can actually collect it at home.

What's next for Teary

Grow the dataset in collaboration with UPJŠ to push five-class F1 past 80%.
Ship the home kit as a real pilot — packaging, cold-chain-free shipping, turnaround logistics.
Expand the keepsake catalog (jewelry, glassware, prints) and launch a made-to-order pipeline from sample → dendrite scan → etched object.
Explore longitudinal tracking: what does your dendrite look like over time, and what does that say about your health?

Teary · one drop · one story