Common Sense Logic
Precise software for decisions where error is expensive.
We model the structure of a problem before choosing a method. Everything we ship is validated on data it has not seen and runs on your own hardware.
The simplest model that explains the data wins.
Structure first
Most problems have structure worth exploiting. A model built on it is smaller, faster and easier to audit than a generic one.
Out-of-sample validation
A model ships only if it outperforms the current method on data withheld from fitting.
Local execution
Our software runs on your machines. It requires no accounts and sends no telemetry.
Two Windows applications, both free to start.
HESI Universal Archiver
RampPlanWhat we build for clients.
Custom mathematical models
A compact model of your specific problem, fitted to your data. It runs in minutes on ordinary hardware, and every output can be traced to its inputs.
Process
- Benchmark the method you use today.
- Withhold the most recent data.
- Fit and calibrate the model, then score it once on the withheld data.
- Deliver software that runs without a server.
- Forecasting. Each customer's history is modeled individually. Sparse histories borrow strength from similar customers.
- Ratings and probabilities. Ratings are estimated jointly, so opponent strength cancels out. Predicted margins map to calibrated probabilities.
- Scheduling. The full week is solved as a single network, and an independent check verifies every plan.
- Risk scoring. Each score is reported with the factors that produced it.
Compression licensing
HESI Professional and Enterprise licenses, and paid pilots on your own data.
Site planning
We source current imagery of your venue and configure RampPlan around it.
Private AI deployment
Language models run on your hardware, on a network with no route to the internet.
Model evaluation
We measure whether a model outperforms the one it replaced, with confidence intervals.
Describe the decision you need to make.
We'll tell you if a model would help.