Omnibio

OmniBio is a zero-shot cytometry console. A researcher drops a microscopy image into the browser and gets, in one run, a cell segmentation, per-object morphological measurements (area, circularity, eccentricity, nearest-neighbor distances), a written cytological synthesis, and a Python analysis script generated from the observed data and executed in a sandbox to produce publication figures at 300 dpi. The problem it solves: cell image analysis normally requires training or fine-tuning a segmentation model on annotated data, then handwriting measurement and plotting code for every new experiment. Most labs have neither the annotations nor the engineering time. OmniBio removes both steps. It segments with Meta SAM 3.1 in zero-shot mode, arbitrates the result against a classical scikit-image fallback using measured quality criteria, and degrades gracefully, every output carries its provenance, so a fallback never masquerades as a model result. We validated it on the full BBBC038 collection (670 nuclei microscopy images) against the official Broad Institute ground truth: mean Dice 0.838, AJI 0.689, instance F1 0.776, with no training or fine-tuning. The interface is bilingual (French/English).