Our research
SPECIALIZED POLYMER AIA research stack built for polymers, not a general-purpose chatbot pointed at chemistry. It predicts structure-property relationships today, and is built to run inverse design next - grounding every recommendation in the literature and running privately on your data. The payoff is not faster answers; it is the next experiment, reached sooner and likelier to work.
AITargets in, ranked candidates out
State the properties you need; the model searches polymer space for repeat units that hit them, ranked by fit. Hover a candidate for its tradeoff.
Illustrative example: one target set; fit scores are normalized match scores, not measured values.
The research stack
A general model guesses. RePolymer reasons, treating a polymer as one coupled system: repeat-unit chemistry at the base, whole-system reasoning at the top. Four specialized layers, each one you can interrogate, each building on the one beneath it.
Reasons across structure, mechanism, and lifecycle.
The top layer plans the investigation, decomposing a question into structure, mechanism, and lifecycle sub-problems and routing each to the evidence and models beneath it.
Literature, patents, and technical reports, evidence-linked.
Built to ground every claim in a retrieved source, linked so you can follow the reasoning back to the evidence it rests on.
Glass transition, modulus, solubility, degradation.
Machine-learned models predict the properties that decide a polymer: glass transition, modulus, solubility, and degradation, reported as honest ranges rather than false precision.
Repeat-unit chemistry, MW distribution, crystallinity, processing history.
At the base sits a rigorous representation of the polymer itself: the structural and processing detail that everything above is built on.
Structure → property
Available nowPolymer properties emerge from repeat-unit chemistry, molecular-weight distribution, crystallinity, and processing history, not from a single SMILES string. The stack encodes those hierarchies, so predictions reflect how polymer scientists actually reason about structure.
That grounding lets it do more than predict: it is built to propose which experiments to run next, so teams spend their effort on candidates with a real chance.
PolyProp runs this property prediction live. Sign in to try it on your own repeat unit.
[ –CH2–CHR– ]n
Illustrative example
These specific numbers show the kind of output the intended structure-to-property capability returns. They are not a live prediction. Today PolyProp predicts fiber-reinforced composite properties from micromechanics; single-repeat-unit property prediction is in development.
Bar length is each property's percentile across known polymers, so the fills stay comparable even though the units are not.
// illustrative example, not a live prediction
Capabilities
Two commitments shape the whole stack: ground every recommendation in a source you can check, and weigh every polymer design against its full lifecycle.
Scientific claims without citations are opinions. RePolymer retrieves and synthesizes peer-reviewed literature, patents, and technical reports, linking every recommendation back to its source so you can verify, extend, or challenge the reasoning.
This is how AI should work in polymer science: augmenting expert judgment with tireless recall and structured synthesis, not replacing it.
A polymer optimized for strength but impossible to recover is a liability. In the same frame as thermal and mechanical performance, the stack is built to weigh end-of-life pathways: mechanical recycling, chemical depolymerization, compatibilization, and biodegradation.
Sustainability is not a separate product line. It is a design constraint, weighed from the first candidate rather than bolted on at the end.
Collaboration
The hard part of polymer science is where fundamental research meets industrial scale-up. We want to work with universities, national labs, and materials companies to test our models against real formulations, real batches, and real constraints.
Your work stays yours. Proprietary formulations and results stay private, and we never train a shared model on your data. Polymers are where we start; broader materials intelligence is where this is headed. Whether you are advancing sustainable polymers, biomaterials, coatings, or agrochemicals, these problems are too important, and too valuable, to solve alone.
Get started
Request access to run our models on your own polymers, or reach out to scope a research collaboration. Your work stays yours; we never train a shared model on your data.