Interactive math items, authored by AI, scored by a solver.

LidNote is a geometry engine that automates the creation of interactive math items for assessment and courseware providers. An author describes the item in a sentence. An agent constructs it as a live figure. A solver scores the student's interaction — the agent builds and the engine marks, ensuring outputs are always reliable.

Illustrated example: "Construct a dodecahedron with its displaced icosahedron dual. Success: the icosahedron repositioned as the dodecahedron seed."

solver: solution condition met solver: solution condition unmet

The most expensive item type in assessment

Technology-enhanced items — drag a point, complete a geometric construction, manipulate a figure — test understanding in ways multiple choice cannot, and they score automatically at scale. Today, every one of them is made by hand.

Built by developers

Simulation-grade items are bespoke engineering: custom rendering and hand-written score logic, weeks per item pattern, maintained for years.

Or bought as labor

Content-service firms build interactives as billable hours. Single item cost is steep and this expense scales linearly with catalogue size.

AI authoring stops at text

AI item generation drafts multiple-choice, cloze and distractor item types across the industry. No tool authors complex interactive item types.

The gap the author describes the item — the AI builds it, the scoring is guaranteed. No platform solves for this.

One prompt in. A live, auto-scored item out.

Author prompt
“Construct a dodecahedron with its displaced icosahedron dual. Success: the icosahedron repositioned as the dodecahedron seed.”
Result: a fully-interactive figure — live geometry, draggable in 3D, scored by the solver on check.
  • The agent builds the figure

    It works with our typed operation language — never freeform generated code, ensuring all outputs are reliable and editable.

  • The solver scores the student

    Every student interaction is semantic scene state. Correctness becomes a deterministic query. Adaptive feedback occurs in an instant.

  • Embeds directly into your product

    Your house style, your platform, your item bank, your control — delivered as an API, an MCP server, or an embeddable SDK.

How it works

Agent operation spec

The model plans the item via typed, deterministic operations against the engine — no freeform code generation.

Geometry kernel & solver

A scene graph with expression-linked parameters and a constraint solver holds one source of mathematical truth.

Interactive layer

Students drag, rotate and construct in 2D and 3D. Every gesture lands as semantic scene state, readable by solver and agent alike.

Authoring & delivery

Solution criteria, adaptive feedback from live scene read-back, and export into the platforms where items are delivered.

“Scoring is a solver query. The engine represents, renders, and verifies what the models draft.” The design policy behind LidNote

Built to be licensed

Assessment and courseware platforms have licensed deep math components for decades — calculators, equation input, subject-specific interactives — capabilities that are necessary but not differentiating. We act as the component vendor, providing the industry with a solution to its most expensive item category.

Deep, and never finished
A geometry kernel, constraint solver, agent interface and reliability eval corpus — years to parity for an in-house build, with no end state.
Off your critical path
Platforms compete on delivery, psychometrics and integrations. The kernel is our entire company; it improves every quarter without your roadmap having to pay for it.
Maintained for you
Keyboard-navigable, screen-reader-meaningful items, maintained against evolving conformance requirements, device matrices and interop.
Neutral by design
We are infrastructure. We are not distribution. Competing with customer platforms for learners, authors or content would collapse our own market.
Hosted APIMCP serverEmbeddable SDK

Now signing design partners

We're working with a small number of assessment and authoring platforms to co-develop the first agent-authored item patterns against a shared reliability benchmark — ahead of general availability.

Platform access is currently reserved for pilot licensees. If you're an investor, an assessment platform, or a publisher with an interactive catalogue — we'd like to talk.