aime™ unveils Curriculum Foundry™ — a provenance-grounded engine for the world's curriculum graph

aime today introduced aime Curriculum Foundry™, the engine that builds the curriculum graph the rest of the aime stack teaches from. The Foundry ingests official government curriculum documents and compiles them into a structured, provenance-grounded graph designed to be audited line by line by the ministries and exam boards that own the source material.

Why the curriculum graph is the bedrock

Everything a learning company produces downstream — lessons, assessments, adaptive pathways, teacher tooling — inherits the quality and the errors of the curriculum model beneath it. In practice, three properties are almost always missing from existing curriculum knowledge graphs. Provenance: alignments are asserted but nobody can point to the exact line of the official document a node came from. Completeness: gaps are silent because nothing proves every teachable unit in the source actually appears. And scale without quality loss: curating one country by hand is expensive, and curating every country, every grade and every subject by hand is not a project — it is a wish.

aime Curriculum Foundry™ is built around a single design rule: quality is a property of the pipeline, not a hope pinned on review. A graph is created correctly because the engine cannot emit one that fails its gates.

“A graph is not created correctly because someone checked it. It is created correctly because the pipeline cannot emit a graph that fails its gates.”

Provenance is computed, not asserted

The Foundry's first hard gate is provenance. An extracted node cannot enter the graph unless its true source is computed and verified against the official document — not taken on faith, not inferred from a model's summary. Ungrounded nodes are rejected at admission. This is what allows a ministry to open the graph on any node and land on the exact line of their own document it came from.

Quality enforced by construction

Each curriculum the Foundry compiles is checked against a set of quality themes — coverage against the source, correct typing, grounded provenance, pedagogically sound decompositions, structural coherence, local context and inclusion. The themes are not a post-hoc checklist. They are built into the engine as gates, so a class of defect found in one country is fixed for every country the engine will ever build.

The Education Council — deliberative, adversarial, in-country

Every compiled graph is evaluated by an Education Council — a multi-agent review body that examines the graph against the government's official documentation before it reaches a human. The council reviews from multiple specialist lenses in parallel, then challenges its own findings adversarially and reconciles the remainder into a single verdict with a full audit trail. It makes the human reviewer's job tractable; it does not replace them.

A deliberate design choice runs through the Foundry: the council rates the appropriateness of aime's own generated content — the decompositions, scaffolds and localisation passes — and never rates whether a country's curriculum is "good." The Foundry's job is to represent the official curriculum faithfully, not to grade the ministry that wrote it.

Risk-based human sign-off

Nothing ships until an in-country teacher signs off. Because humans cannot scale to every node of every grade of every country, review is risk-based: the engine auto-clears the majority while a human reviews the flagged minority — novel, low-grounding, low-confidence and culturally-sensitive nodes. A human always stays on the culturally-sensitive class, which a model monoculture cannot self-audit.

Why this is defensible

The individual capabilities the Foundry uses — parsing, extraction, prerequisite graphs — are commodities. What is not commodity is the integrated loop: provenance-gated construction that refuses ungrounded nodes on messy real government documents; an adversarial education council operating at node admission rather than as post-hoc scoring; and a cross-country graph that compounds as each new curriculum is reviewed. The specific gate implementations, council architecture, prompts and grounding-repair strategies are aime's proprietary IP and are shared with partners under NDA.

Status

The software pipeline is complete and tested end-to-end. A pilot curriculum has been taken through multiple council rounds to a fully grounded graph, with all gates passing. Additional countries are now entering the build queue. The Foundry is moving from pilot into a country-by-country content build.

“Curriculum modelling has been a hand-curation problem for as long as ed-tech has existed. We are turning it into an engineering discipline — provenance computed, quality gated, only our own additions judged, and a teacher signing off on the flagged minority. The output is a curriculum graph a ministry could audit line by line. That is the bedrock every downstream learning product deserves.”

About aime

aime builds the operating system for educational intelligence — the foundational infrastructure layer that future education systems will run on. aime's stack combines structured curriculum knowledge, pedagogy-aware reasoning, compact education-tuned models, agentic orchestration and offline-capable deployment, and is designed for ministries of education, universities and national school systems.