A
Equation extraction
Parse equations and symbolic structures directly from scientific literature — typed, unit-aware, and traced back to source.
Ekaform extracts equations from scientific literature, connects them into a calibrated knowledge graph, and compiles them into executable, reproducible simulations. Terraform did it for cloud infrastructure. We do it for science.
Built for the next generation of AI-native science.
Six stages. One compiler for scientific knowledge. Each step is auditable, reproducible, and traceable back to its source.
Ingest open-access literature, technical reports, or your internal corpus.
Parse equations, parameters, and symbolic structures from raw documents.
Connect variables, assumptions, and physical relationships into a typed knowledge graph.
Optimize parameters against real observations through differentiable data assimilation.
Run differentiable, reproducible simulations from the compiled scientific graph.
Deploy as an API, embed in agents, or version it like any other production system.
Parse equations and symbolic structures directly from scientific literature — typed, unit-aware, and traced back to source.
Connect models, variables, parameters, and physical relationships into a queryable, versioned graph.
Calibrate the graph against your experimental data. Bayesian inference, gradient descent, and uncertainty bounds — built in.
Compile to a differentiable runtime. Run on CPU, GPU, or distributed clusters — same code, same results.
Composable libraries of executable scientific knowledge. Import a model the way you import a package.
Every parameter traces back to the paper, equation, and dataset that produced it. Built for regulated science.
Fifty million scientific papers contain the equations to simulate molecules, climates, markets, and materials. They sit, unstructured, inside PDFs.
Every team that needs a model rebuilds it from scratch — re-deriving equations, re-coding solvers, re-calibrating parameters someone else has already published. The cost of that reconstruction is measured in years, careers, and entire research programs.
Ekaform converts scientific knowledge into machine-readable, executable infrastructure — a programmable substrate for the AI era.
Predict PK/PD, toxicity windows, and translational outcomes from literature plus your preclinical data.
Compose differentiable atmospheric, oceanic, and ecological models from the published canon.
Predict stability, degradation, and structure-property relationships from the literature you already cite.
Compose macro and micro models with provenance — every coefficient traceable to its source paper.
Give your agents access to executable scientific knowledge — not just text.
Sovereign, on-premise scientific simulation built from your own classified corpus.
Ekaform exposes scientific models as first-class software primitives. Import, compose, calibrate, and simulate — from a single SDK.
# compose a model directly from the published literature from ekaform import load, calibrate model = load("pk_pd/translation/v4") # calibrate against your preclinical observations model.calibrate( data="./preclinical/mouse_cohort_2025.csv", method="variational", ) # run a calibrated simulation prediction = model.simulate( dose=12.5, subject="human/adult/f", horizon_hours=72, ) prediction.trace() # → equations, papers, parameters, uncertainty
Ekaform is in private alpha with a small group of research and infrastructure partners. If you're working on AI-native science, we want to hear from you.