Models

Turn raw broom and mop sensor traces into labelled activity phases: strokes, corners, lifts, pauses. Run it before you look at any dry-dusting trace by hand.

Predicts shelf-stability class for aqueous surfactant formulations. Use it before you commission a physical stability study, to screen the candidates in silico first.

Rank concepts against a synthetic consumer panel before you recruit a real one. Read it as a pre-screen: rank-order, not verdict.

Reads a folder of raw sensor files and works out what they are: rate, axes, participant, arm. Auto-invoked whenever the agent recognises an upload, so you rarely call it yourself.

Estimates cost per unit for a candidate formulation off the validated BOM ledger. Use it to sanity-check a formulation before it reaches a costing review.

Predicts percentage soil removed for a cleaner-substrate-soil combination. The first thing to run when someone asks whether a new formulation actually cleans better.

Scores visible streaking and dried residue after a wet clean. Use it whenever the complaint is how the floor looks once it dries, not how clean it is.

Flags formulations likely to dull, etch or haze a sealed finish. Run it before any consumer-facing surface claim. It is the safety gate, not the performance one.

Estimates how much performance a formulation loses as water hardness climbs. Use it when a market complaint maps to a hard-water region.

Solves the spray plume for a trigger geometry: coverage, droplet size, wall deposition. Use it when the question is where the product lands, not what it does once there.

Early draft. Estimates how long a scent stays detectable on a hard surface after cleaning. One US panel behind it, so treat every number as directional.

Fits a flow curve and reports the viscosity profile across shear. Use it to catch structural breakdown early, before a stability study moves.