Squirrel AI
one-to-one adaptive learning driven by fine-grained knowledge decomposition
Squirrel AI, a one-to-one adaptive learning system deployed in China, has supported more than 24 million students and set a Guinness record in September 2024 of 112,718 participants in an online math lesson, with mastery rising from 41.6% to 85.1% in synchronized learning.
Key points
- One-to-one adaptive learning through fine-grained decomposition of knowledge points.
- Proprietary system with a knowledge graph and knowledge tracing.
- More than 24 million students, a Guinness record of 112,718 participants, mastery from 41.6% to 85.1%.
- Evidence B, confirmed status (Time Best Inventions 2025, certified record 2024).
Objective
Give each student a plan and one-to-one tutoring generated by the system, with efficiency higher than mass teaching, to run a network of centers and tablets at the scale of tens of millions of students in China.
The deployment
Squirrel AI breaks each subject into thousands of fine-grained knowledge points, then diagnoses what the student masters and builds an individualized path. The system delivers exercises, explanations, and remediation one-to-one, continuously adapted to the answers. It is distributed through physical centers and tablets in China. In September 2024, Squirrel AI ran an online math lesson synchronizing more than 112,000 students, certified by Guinness World Records.
Results Proof B
Scale figures certified by Guinness World Records (September 2024 event) and reported by Time (Best Inventions 2025) and the World Economic Forum; learning gains documented by an academic study. Concordant sources, mostly official or third-party, hence evidence B.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- plateforme Systeme adaptatif proprietaire Squirrel AI engine that diagnoses the student's mastery and generates the one-to-one path, exercise by exercise
- outil Graphe de connaissances par matiere decomposition of each subject into thousands of fine-grained knowledge points, built by the education team
- outil Modele de maitrise (knowledge tracing) continuous estimation of what the student masters, calibrated on the learning behaviors of more than 24 million students
- infra Tablettes et centres physiques distribution network: centers in more than 100 Chinese cities and dedicated learning tablets
How it runs, concretely
For ops teams-
1Diagnosis AI
The system assesses what the student masters across fine-grained knowledge points.
-
2Individualized plan AI
A path of exercises and explanations is generated to fill the detected gaps.
-
3Adaptive remediation AI
At each answer, the system readjusts the difficulty and the next content.
-
4Human support data team
In the centers, supervisors sustain motivation and handle cases outside the model.
The estimated mastery state of each knowledge point. If the diagnosis is wrong, the path proposes miscalibrated exercises and the promised efficiency collapses.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- fine-grained decomposition of the curriculum into knowledge points
- history of student answers to estimate mastery
- sufficient data volume to calibrate the adaptive model
Org prerequisites
- education team to build the knowledge graph
- data science team for knowledge tracing
- distribution network (centers, tablets, or app) and compliance for minors' data
Possible stack
- adaptive learning engine (proprietary or platform)
- knowledge graph per subject
- mastery model / knowledge tracing
The plan, step by step
- Step 1Have the education team build the graph of fine-grained knowledge points on a pilot subjectDeliverable: Knowledge graph validated by teachers
- Step 2Produce the exercise bank tagged on the graph and trace each student answerDeliverable: Instrumented exercise platform, first answer data
- Step 3Train the mastery model (knowledge tracing) on the accumulated answersDeliverable: Mastery diagnosis evaluated against reference tests
- Step 4Close the adaptive loop: generate the individualized path (next exercise, remediation) from the diagnosisDeliverable: Adaptive engine piloted on a group of students
- Step 5Add human support (motivation, cases outside the model) and compliance for minors' data, then extend to other subjectsDeliverable: Full setup measured on before/after progress
First step: Break a subject into fine-grained knowledge points and instrument the answers, before connecting the adaptive engine.
The same pattern, proven in other industries
Taobao
Real-time product recommendation with deep learning, served across hundreds of scenarios in the app
Amazon
Item-to-item collaborative filtering product recommendation engine, served across every touchpoint
ASOS
catalog-scale deep learning product recommendation, plus a conversational style assistant
Sources
- S1 Squirrel Ai Learning Sets Guinness World Record for the most users to take an online mathematics lesson in 24 hours (PR Newswire) Interested party archive pending
- S2 Squirrel Ai Intelligent Adaptive Learning System: The Best Inventions of 2025 (TIME) Established press archive pending
- S3 The Squirrel AI Adaptive Learning System Accompanying Millions of Children in Their Growth (ResearchGate) Secondary archive pending
- S4 This AI tutor could make humans 10 times smarter, its creator says (World Economic Forum) Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.