LANVAR
← The shelf

stem cell classifier

Did its thing

A classification model for stem cells that took first place at PharmaHacks — at the same fair that ignited the whole ML obsession a year earlier.

✳ 1st place · PharmaHacks PythonML 2023
✳ How it actually works
THE CLASSIFIER, RUNNING one cell at a time — three lineages on the board CELL SAMPLES unlabelled INPUT morphology + intensity HIDDEN learned features OUTPUT softmax over 3 LINEAGE CALL NEURAL CARDIAC HEPATIC VERDICT — classifying… NEURAL · 96% HEPATIC · 88% CARDIAC · 93% CARDIAC · 93% CONFIDENCE the wave decides — the lamp only announces it
FIG 1 · the classifier, animated — one cell in, one lamp out
labelled cells the training set feature engineering what signal matters classifier trained to state-of-the-art cell-type verdict the winning output validation tune until it earns it PharmaHacks · 1st place · where the whole ML obsession started a year earlier
FIG 2 · the machinery — cells → features → verdict
The story

PharmaHacks is where the lab’s machine-learning arc started — first as a wide-eyed participant, then a year later back with Aman Dalmia to win the whole thing.

The challenge: build a state-of-the-art stem cell classification model. The work was the unglamorous core of applied ML — wrangling the dataset, engineering features, tuning and validating until the model earned its accuracy. First place, and proof that showing up again beats showing up ready.

Next on the shelf microbiome classifier →