Designing delivery for the next generation of medicine.
Kosha Sciences is building an integrated platform — AI-driven design, automated synthesis, and high-throughput screening — to accelerate the discovery of lipid nanoparticles and the therapies that depend on them.
We are at the start of a remarkable period for medicine. The tools to edit genes, program cells, and tailor therapies to a single patient are, for the first time, within reach. The constraint now is not what we can imagine, but what we can deliver — safely, precisely, and at scale.
The opportunity: intelligent design for drug delivery.
At Kosha Sciences, we are pioneering intelligent design for the next generation of drug delivery and gene therapies. Our mission is to rapidly accelerate the development of next-generation lipid nanoparticle (LNP) technologies — delivery vehicles that protect and transport delicate RNA molecules directly to specific cells in the body.
LNPs played a critical role in the rapid development and deployment of mRNA vaccines during the COVID-19 pandemic, showcasing their safety and effectiveness as a delivery platform. Building on this success, we believe the most promising therapies of the next decade — gene editing platforms, RNA therapeutics, and other precision medicines — will similarly depend on highly effective delivery systems to reach the right cells at the right time.
Today, developing these systems is slow, costly, and largely driven by trial-and-error — which limits how quickly life-changing treatments can reach patients.
Our approach: an integrated platform, grounded in first principles.
Kosha is building a state-of-the-art integrated platform that combines AI-driven design, automated synthesis, and high-throughput experimental screening. Our system dramatically speeds up the discovery of novel ionizable lipids, LNPs, and molecular binders — reducing development costs, shortening timelines, and enabling therapies that currently may be out of reach.
Core to our approach is a belief that better models of delivery come from encoding parsimonious descriptions of the underlying chemistry, biology, and physics — how a molecule crosses a membrane, escapes an endosome, finds the right tissue, and engages or evades the immune system. This guides which representations to use, which priors to encode, and where to apply relevant constraints. The result is models that are more data-efficient and more likely to propose candidates that translate from in silico prediction to experiment, and ultimately to the clinic.
“The ceiling on these models isn't the architecture. It's the quality of the data they're trained on, and how faithfully our representations reflect the science.”
Parsimonious models need clean signals to latch onto — and that starts upstream of the model itself, in how experiments are designed and run. We've built workflows that run at high throughput without sacrificing experimental rigor: consistent protocols, tight controls, and endpoints that measure what we actually care about. The result is data that captures a cleaner, more consistent signal of the underlying dynamics.
Stay in the loop.
Occasional notes on what we're learning about delivery, models, and data.