Foundation models we've trained and evaluation benchmarks we've built, published openly, each one traceable to a paper or preprint.
A codon-level equivariant language model for mRNA: the representation underneath expression prediction, stability assessment, generation, and therapeutics design in the co-scientist.
A plug-in affinity and pose benchmark that treats historical public protein-family support as the independent variable: family-disjoint controls, an external low-support target, and the family generalization gap alongside the headline number.
Every entry carries its review status in the open: peer-reviewed where it has been reviewed, preprint where it hasn't. Never an internal benchmark alone.
Architecture, training data, and evaluation are public and reviewable, not just asserted.
Every prediction that touches mRNA representation traces back to the model that produced it.