Blog
How this is built, and what it measured.
Every post says how the thing it describes was checked, and states plainly what it does not claim — including the evaluation where our own numbers came back worse than we wanted.
- GroundingFor institutionsThe difference is not the model. It is what the model is allowed to read.A general assistant answers from the internet. A course-grounded tutor answers from the material your instructor published, cites the page, and clearly labels anything it explains beyond that material.7 min readRead
- ProcurementFor institutionsThe demo is not the product. These are the questions that find out.A procurement checklist for AI tutoring, written from the side of the table that has to answer it — including the questions we answer badly.9 min readRead
- PolicyFor institutionsBanning the tool does not remove the tool. It removes the evidence.Blanket AI bans move usage off institutional systems where nobody can see it. A layered policy keeps it visible and lets the register change by context.8 min readRead
- ProductFor institutionsA picture a learner cannot move is a slide with extra steps.An image model draws something that looks like a circuit. A parameterised template draws the circuit, and the learner can change it.6 min readRead
- EvaluationFor institutionsA number we would rather not publish, published.Our own evaluation found grounded STEM answers far weaker in Spanish and Polish than in English. Publishing the gap is more useful to a buyer than hiding it.5 min readRead
- Study ToolsFor learnersFour questions that separate study tools that teach from tools that answer.What actually separates AI study tools — where the material comes from, whether answers are checkable, and whether a tool tells you when it has gone beyond your notes.6 min readRead
Comparing us with something else?
Each comparison names at least one thing the other product does better, because a comparison we win on every row is an advertisement.
See the comparisons