RaRaMa: structure-only prediction of taste, protein function and process outcome for food and ag ingredients, computed from first principles, not trained, and scored on your own held-out data

What it is. RaRaMa predicts what a food or ag ingredient will do, from its structure. We have a specific format where you would send a protein sequence, a molecule, or the settings from your process, and get one clear report back in minutes..

taste and off-notes, aroma, melting point, gelation, foaming and emulsifying, solubility, protein quality, allergen family, and whether it survives your cook, extrusion, spray-dry, membrane, ferment or high-pressure step.

Why it matters. Making and testing a new candidate takes weeks and real money before you learn anything. RaRaMa gives you the read in seconds, while the sample is still being made, so you only pay to test the candidates worth it, and you get a head start on the identity and safety work regulators ask for.

How it works. Every answer is computed from the structure and the known behaviour of water, and comes with the plain reason behind it. The same input always gives the same answer, you can check it, and a brand-new molecule is read as reliably as a familiar one. When something is outside what the engine knows, it says so instead of guessing, thus being a value to R&D everywhere.

What it is built on. This is not one or two lucky results. The engine has been scored on held-out data across thousands of taste compounds and proteins, spanning precision-fermentation, pulse and oilseed, dairy and egg proteins, and industrial enzymes, plus full process runs. I've been testing it myself for a long time. And with representative, independently checked results: melting point within about 3 degrees inside a known protein family, intact mass within about 0.1%, allergen family correct 88% of the time against the WHO/IUIS reference, and a 0.95 correlation with an outside contracted taste lab. On taste it reads (again) the structure, not the label, telling apart molecules that differ by only a single bond.

Where it stops. Absolute gelation and emulsion strength take one bench measurement to lock in, and before that you get the order and the risk flags. For a protein with no single melting point, like wheat gluten, it refuses to invent a melting temperature and tells you to measure it, while still returning the allergen family, the structure and the rest of the read. It front-runs the lab; it does not replace your final confirmation.