BlueberryML Ltd
AI infrastructure
for education.
Six composable AI services — grading, content generation, personalisation, reporting — available via API or as a full platform. Use our intelligence with your data, or let us be your trusted data custodian.
- 100,000+
- Students
- 100+
- Schools
- 20,000+
- Activities
- 99.9%
- Grading Accuracy
Six composable services
Each service is a tool. Together, they form an autonomous learning pipeline.
Use one. Use all six. Call them via API, or deploy as ClassGrade — the full platform.
Curriculum Embedding
Any curriculum — UK, Kenya, India, Japan, Italy — mapped into a unified learning outcome space. Every topic, concept, and prerequisite relationship structured and queryable.
curriculum.embed() — Ingest a curriculum. Returns a structured knowledge graph with prerequisite edges.
Assessment & Input
Dynamic forms, exam builder, worksheets, voice tasks. Configurable per school, per publisher, per curriculum. Any assessment format becomes structured data.
assess.create() — Configure assessments against any curriculum node. Written, spoken, and interactive formats.
Analysis & Grading
OCR for handwritten work at under 0.1% failure rate. Voice recognition at 99.9% accuracy. AI rubric scoring. Student work graded instantly and mapped to learning outcomes.
grade.submit() — Submit student work as image, audio, or text. Returns scored results mapped to learning outcomes.
Knowledge State
Each student's results position them in the learning space. Mastery, developing areas, and gaps identified. Prerequisites tracked. A live knowledge graph per student.
student.state() — Query a student's knowledge graph. Returns mastery probabilities, gaps, and recommended next nodes.
Adaptive Content & AI Tutor
150+ activity templates, 20,000+ exercises generated from curriculum. Plus an AI tutor — voice and chat — adapting in real time to each student's knowledge state.
learn.next() — Returns the optimal next activity or initiates a tutoring session. Model and difficulty selected automatically.
Reporting & Analytics
School dashboards, teacher reports, parent analytics, publisher-level insights. Real-time, automated, zero admin overhead.
report.generate() — Reports at student, class, school, or network level. Structured data or formatted output.
Two ways to work with us
Your data or ours. Same intelligence.
For publishers & assessment bodies
API
You hold the data. We provide the intelligence. Integrate our grading, content generation, and personalisation services into your own platform via API. Your data never leaves your infrastructure.
- —Composable microservices — use one or all six
- —Your data stays in your environment
- —Model-agnostic — we orchestrate the best AI for each task
- —White-label ready
For school groups & franchise networks
Platform
ClassGrade is the full bundled product — all six services composed into a complete school platform. BlueberryML acts as your GDPR-compliant data processor, handling children's learning data, biometric traces, and assessment records under strict regulatory obligations.
- —All six services, ready to deploy
- —BlueberryML as trusted data custodian
- —GDPR Article 28 compliant — DPA with every client
- —Zero training on client data by any foundation model
Why this matters
The AI gets better every quarter.
The data is what's protected.
Foundation model providers will always build better models. But they cannot access children's learning data directly — GDPR and child protection regulation prevent it. BlueberryML sits in that gap: orchestrating the best AI against real, permissioned educational data.
GDPR & data sovereignty
Whether you hold the data or we do, the regulatory framework is enforced. Children's data never reaches foundation model training pipelines.
Model-agnostic
Anthropic for reasoning, open-source for on-prem, smaller models for latency. The AI improves every quarter from every provider. We orchestrate — you don't lock in.
Agentic composition
Each service is a tool. Bound together agentically, they form autonomous workflows — curriculum in, personalised learning out. Or call them individually for exactly what you need.
Children's data as the moat
Real learning traces, biometric data, handwriting, voice, interaction patterns. Data that GDPR makes inaccessible to big AI directly. Accessible — safely — through BlueberryML.
Market traction
Live with 100,000+ students. Not a pilot.
ClassGrade — the full platform mode — is deployed daily across a national school network. Full grading throughput, teacher reporting, and student activity generation in production.
Publisher API integrations in progress across multiple markets and curricula.
Read the case studies →Chrysalis Learning
Full platform deployment across a franchise network. Daily teacher usage, automated grading, and reporting at scale. BlueberryML as data custodian.
The opportunity
Every publisher needs this. None of them will build it.
Curriculum publishers have content. They lack the AI infrastructure to make it adaptive, auto-graded, and personalised. Building it in-house means solving GDPR, child safety, model orchestration, OCR, voice recognition, and knowledge graph computation — simultaneously.
Or they call one API.