Four layers — experiment data, design models, the reagents a run consumes, and the capacity to execute it — wired into one loop. Owning the loop is the difference between selling reagents and running experiments.
Every project produces two kinds of output: the answer you asked for, and everything the bench learned on the way. Most suppliers keep the first and discard the second.
A sequence on its own does not express. Codon usage, signal peptide, tag position, host and induction window decide whether you get protein or inclusion bodies. The models propose that combination and rank it before anything is synthesised.
The layer your experiment actually touches. Designed against an assay rather than a catalogue slot, then released under a quality system that survives audit.
Our own laboratory executes: expression, purification, assay, release. When volume exceeds one bench, qualified partner facilities absorb the overflow under our methods and our quality system.
Four steps per round, including the rounds that fail. The failure is the data.
Model-proposed constructs, ranked by predicted yield and soluble fraction.
Gene synthesis, cloning and transformation — in house or through qualified partners.
Expression matrix at bench scale: titre, soluble fraction, purity by SDS-PAGE.
Every result — including the failures — updates the model for the next round.
The loop is the product. A reagent shipped without its assay data is a commodity; the same reagent shipped with the data that shaped it is a head start.
AI claims are cheap. Being specific about the boundary is what makes the rest credible.
We do not sell software licences or a foundation model. The models are internal tooling that makes our experiments and reagents better.
We do not take on every assay. We take work that extends the tool stack — if it teaches the platform nothing, we will say so.
We do not sell hardware, plate readers or closed cartridges. We design the experiment for the platform already on your bench, and route the work when the bench is not yours.
Send an assay that is not behaving. We will tell you which layer is failing — and whether it is worth fixing with a tool, a protocol, or a different approach entirely.
