CocoRobo
SMART Classroom
CocoRobo builds AI-native learning environments on the SMART principles: helping teachers see how students think, and helping students keep understanding, judging, reconstructing and co-creating with AI support.
AI is powerful — but education's real problem isn't efficiency
AI can generate answers, explain concepts, design exercises and grade work. But if students just finish faster, teachers just hold more data, and classrooms just get busier — technology hasn't really changed education.
Faster to finish — but deeper in understanding?
Writing an essay with AI doesn't mean knowing whether its claims hold; solving a problem doesn't mean knowing why you were right or wrong.
More interaction — but is thinking truly visible?
Many clicks and submitted answers don't add up to learning if the thinking process is never seen, compared or discussed.
Stronger AI — but do people keep judgment?
The real risk isn't that students can't use AI, but that they finish faster while growing more fragile in deep understanding and knowledge responsibility.
A smart classroom is first of all a high-quality learning environment
A room with big screens, tablets, an AI assistant and a data dashboard is not yet a smart classroom. A real smart classroom is defined by how learning happens — not by its devices.
Build a better learning environment
Let students' ideas, questions, errors, evidence and artifacts enter a shared space, becoming resources the class can build on.
Cultivate more mature learners
Students don't just use AI to finish tasks — they question, judge, revise, take responsibility, and co-create with peers.
Let AI scaffold, not substitute
AI should not skip students' thinking; it should walk with them through difficulty, support teacher judgment and protect learners' chance to reason.
Five mechanisms that make a smart classroom real
SMART is not a feature checklist but a set of learning-quality criteria — helping us judge whether technology truly makes learning visible, teaching diagnosable, scaffolding precise, understanding deeper and learning co-created.
Sharing
Bring students' ideas, questions, drafts and errors into the classroom's shared space.
“Good sharing doesn't display results — it surfaces understanding.”
Measurable
Turn the key evidence in learning into judgments teachers can read and act on.
“Not more data — sharper teaching judgment.”
Adaptive
Give just-enough scaffolding at the crucial point, instead of handing over the answer.
“Not thinking for the student — helping them keep thinking.”
Reconstructive
Reorganize understanding through explanation, comparison, revision and transfer.
“Not new formats for content — a real upgrade in understanding.”
Team
Let students discuss, compare and revise around a shared goal toward a better joint outcome.
“Not grouping — thinking a problem through together.”
One suite, supporting the full SMART learning loop
The CocoRobo suite is not a set of scattered tools but an AI-native learning environment designed around the SMART principles — spanning in-class interaction, self-paced study, collaborative projects, classroom analytics and teacher AI-workflow design.
Not slides put online, but traditional courseware turned into a classroom that is interactive, diagnosable, scaffolded, upgradable and co-created — with orchestration (CIA: content · interaction · analysis) helping teachers read the whole class instantly.
- ✓One-click PPT → HTML courseware — Upload a PPT and it's auto-converted to an HTML online lesson (embedded video stored in the cloud); teachers assemble it like editing online slides, adding pages as they go — no re-prep, keeping existing habits.
- ✓AI Agent teaching assistant — Role-play, Socratic dialogue, concept explanation and classroom interaction make lessons more alive.
- ✓Embed diverse interactive tools — Within one page add multiple-choice, fill-in (with AI auto-grading), Q&A, polls and photo capture — and embed agents, workflows or H5 built in CocoFlow, with whole-class thinking shown live.
- ✓PIN co-screen + instant diagnosis — Students join on-screen with a PIN; once they hit submit, data streams back in real time — teachers see every student, and AI summarizes the whole class (since a teacher can't review each one). With heatmaps, follow/free modes and one-tap Freeze.
- ✓Post-class report (coming soon) — After class, a report is auto-generated from the data collected in class; with 1:1 tablets it captures each student, and falls back to teacher-side information when devices are limited.
CocoClass main chain
- 1ImportCourseware
- 2UnderstandPage & goals
- 3AssembleInteractions
- 4InsightClass state
- 5SummaryReports
- 6DispatchHomework
CocoStudy main chain
- 1InputPaper/practice
- 2RecognizeDigitize
- 3DiagnoseItem + whole
- 4PathPlan
- 5CoachExplain/variant
- 6ProfileUpdate & loop
Not a drilling app, but a personalized learning system with diagnosis as entry, the learning path at its core and portfolio update as the loop (IPO: input · processing · output). Practice becomes a diagnosable, scaffolded, accumulating process.
- ✓Smart quizzing + paper digitization — Single/multiple/blank/open items, OCR import, LaTeX formulas and AI open-response marking — cutting repetitive work.
- ✓Data insight & diagnosis — Auto-detect weak points, separate carelessness / method / concept errors, and give class overviews with teaching suggestions.
- ✓Path first, problems second — Socratic prompts let students find the shared issue themselves, then fill gaps and try variants.
- ✓Self-study loop + portfolio loop-back — Diagnose → plan → learn → practise → assess → improve; results write back to the portfolio and flow to CocoClass.
Not a note app, but a workspace where groups gather materials, raise ideas, accumulate evidence and form solutions around a real problem — moving collaboration from division-of-labour to co-creation, and strengthening socially shared regulation (SSRL).
- ✓Hypermedia board · cards · chat — On an open canvas, build knowledge non-linearly with function cards (topic / phase / task) and content cards (mind / AI / resource / board / camera); group chat supports live collaboration.
- ✓Three reactive agents — Planning / monitoring / reflection assistants give structured scaffolding and externalization tables by phase.
- ✓Proactive agent · Lightbulb — Detects triggers like low engagement or negative emotion and offers timely metacognitive and socio-emotional support.
- ✓From division to co-creation — Renders collaboration dynamics and idea evolution as a living graph for cross-disciplinary PBL.
Research evidence
- ▸In a quasi-experiment with 78 sixth-graders, the agent-equipped group significantly outperformed the control on reflection/evaluation (n=38 / 40).
- ▸More balanced participation: 6/6 members engaged, 85% of task-related talk guided by agent prompts.
- ▸The CocoNote work received an ISLS/CSCL Outstanding Short Paper award; MIRACLE adds multi-agent regulation.
Four design principles
- 1ReduceReconstruct the real lesson
- 2AnalyzeMulti-dimensional insight
- 3VisualizeCharts over text
- 4AgileFast, modular feedback
Move teaching from 'gut feel' to 'observable'. Teachers upload a class recording; CocoView uses speech recognition and multimodal analysis to reconstruct the real lesson and diagnose teaching and learning with visuals like the activity spectrum and problem-chain tree — a dedicated lens for the Measurable dimension of SMART.
- ✓Activity spectrum — Colour the whole lesson by teacher-led vs student-autonomous time — see time allocation and inquiry continuity at a glance, and compare a lesson taught twice.
- ✓Problem-chain tree — Reconstruct the lesson's questions into a Bloom-levelled tree — locate the cognitive gradient and logic gaps, with alerts when a sub-question drifts off the main line.
- ✓Multi-framework, subject-specific — General modules — S-T, word cloud, IRF, scaffolding, LICC, TPACK — layered with subject-specific frameworks for Chinese, Math and Science.
- ✓Teaching–learning dual track — Align the observation track (teaching behaviour) with the interaction track (student responses) to the second, revealing how teaching and learning match — or break.
Not AI taking over teaching, but teachers becoming the designers of the AI-era teaching process — able to build three independent things: conversational Agents, multi-step Workflows and interactive H5. It defines when AI enters, what it does, what it must not do, and when the teacher must confirm.
- ✓Three independent outputs: Agent · Workflow · H5 — CocoFlow builds conversational agents, multi-step workflows and interactive H5 separately — independent and used as needed (H5 is something almost any LLM can generate too); each is embeddable straight into CocoClass and CocoNote.
- ✓No-code · teacher as creator — With no coding background, teachers preset standard tutoring logic (explain → example → guided practice) and build classroom-fit agents and interactive pages in plain language, so the whole class gets consistent guidance.
- ✓RAG against hallucination — AI strictly seeks answers first in the school's own knowledge base (textbooks/notes), staying within the curriculum.
- ✓Boundary Learning · productive friction — Keeps the teacher's judgment and final say — not frictionless automation, but productive cognitive friction.
Four judgments the teacher keeps
- ▸Task-boundary recognition: which steps suit AI, which must stay human.
- ▸Model-boundary diagnosis: where AI output is reliable, where it needs review.
- ▸Pedagogical-responsibility demarcation: the line between AI suggestion and teacher decision.
- ▸Structured configuration: fixing teaching logic into reusable, controllable workflows.
Getting started: create agents, build, self-study, collaborate & review
Five core products, five starting lines: with CocoFlow build your own teaching agent with no code; with CocoClass turn a PPT into an interactive classroom; with CocoStudy run a diagnose–tutor–vary–reassess personalized self-study loop for each student; with CocoNote let students collaborate on an open canvas; with CocoView turn one lesson into a readable teaching-and-learning diagnosis. Switch products above and swipe through each flow.
New app, pick Simple mode
Create a blank app in AI Apps — three types: Simple (Agent) / Advanced (Workflow) / AI Coding; name it, add a blurb, auto-make a cover with the magic wand.
Write the system prompt
Define role, task and boundaries with CO-STAR or Role·Skills·Process·Output·Limits; the magic wand drafts it with AI — always review and refine.
Add a knowledge base and tools
Upload standards and lesson plans to the knowledge base (RAG for grounded answers) and say when to retrieve; bolt on image generation, Bing and Bilibili search, speech and web-fetch.
Test, then pick a mode
Chat-test in the debugger and iterate the prompt; choose a delivery mode — Immersive (cartoon companion, K-9 classes), Card (chat plus workspace) or Conversational (quick Q&A).
Publish and use in class
Publish in one click to get a link and class code; in Advanced mode, orchestrate several agents into a workflow. Then call it inside CocoClass courseware or a CocoNote board as an AI App.
* Advanced mode orchestrates several agents into a workflow; screens are anonymized.
Choose a creation mode
Recommended: upload a local file to build fast from existing courseware — or start from a blank.
Edit the base pages
The system auto-parses every PPT page; tweak text and images, and add title or image-text pages.
Add interactive tools
Within a page add multiple-choice, fill-in, Q&A, AI apps and H5 web pages; CocoAI can help generate the content.
Add more resources
Insert video and other media — upload locally or search directly on Bilibili.
Publish the lesson
After a preview, pick subject, grade and class, set the visibility, and publish to students in one tap.
* Creation modes such as 'From AI' and 'From library' are rolling out.
Test & digitize
Photograph or upload any paper, worksheet or PDF and AI segments the questions. A confidence engine routes by high / medium / low — auto-process, suggest-and-confirm, or hand to human review — so any paper just works.
Multi-layer error analysis
Beyond right-or-wrong: Layer 1 spots the error, Layer 2 tags the misconception (careless / method / concept / expression), Layer 3 maps the whole paper to find the common cause behind many mistakes.
Personalized path planning
The system picks the next move per student: teach the concept first for a concept gap, use Socratic questioning for a method error, or generate variant questions when it is mastered but needs reinforcement.
Guided practice
No more dry explanations. The chat flow offers self-directed and AI-guided modes; when a student stalls it surfaces podcasts, mind maps and memory cards to lower cognitive load.
Portfolio update
Every interaction updates the student portfolio — history, error trajectories, class participation, interaction summaries; mistakes and notes auto-archive into a personal error book and an end-of-term report.
* For junior & senior high; capabilities and UI follow the live product.
Frame the project
On an open board, use topic / phase / task function cards to break one real big question into a workable project skeleton.
Express & build
Students add ideas with content cards — mind, AI, resource, board and camera — arranging, linking and organizing freely.
Peer collaboration
Groups share, negotiate and build ideas, learning through idea conflict; within- and cross-group collaboration are supported.
Instant teacher feedback
Teachers push resource and scaffold cards; students submit; AI dialogue can assist task design and brainstorming.
Data & presentation
The dashboard shows contribution, collaboration and an activity log; one-tap present lets groups show results in an organized way.
* Screens and case are anonymized.
Upload the class recording
Drop in classroom audio or video (MP3 / MP4…), or auto-sync from the recording system; supports demo, everyday and same-lesson comparisons.
Pick subject and template
Choosing a subject auto-recommends a subject-specific framework — Science 'scientific-thinking,' Chinese 'read-think-express / PISA reading,' Math 'TRU Math' — layered on general modules (activity spectrum, S-T, Bloom question types).
Verify the transcript (optional)
Speech is auto-transcribed; for demo or showcase lessons you can review and correct the transcript so later analysis is sharper.
Submit; report in ~15 min
Submit in one click; the server runs speech, semantic and multimodal analysis (a 45-min lesson in about 15 min); the task board tracks status and history.
Read the report — teaching meets learning
A visual report: activity spectrum (teacher / student time, inquiry continuity), problem-chain tree (cognitive levels and logic gaps), S-T and Bloom; the observation track (teaching) and interaction track (learning) align to the second for a side-by-side diagnosis with tips.
* Four principles: reduce · analyze · visualize · stay agile; schools anonymized; case from primary-science lesson-review practice.
From a lesson to growth: how SMART unfolds in real learning
The same SMART principles can land in completely different subjects, ages and lesson types. Each case below is a traceable chain of learning — switch tabs to see how SMART actually happens across different classrooms.
Every case below is distilled from real classroom practice.
Take a primary-school science lesson — a soil water-permeability experiment. This isn't a lesson dressed up by AI, but one where the learning process is genuinely seen, diagnosed, scaffolded, reconstructed and co-created.
Before class: design the environment
In CocoClass the teacher imports slides and embeds an H5 permeability experiment; the system generates interactive questions and observation tasks from the learning goals.
Express: thinking enters the shared space
Students predict permeability rates for different soils and submit their reasons; the whole class's ideas appear live on the board.
Diagnose: see the shared misconception
The system finds many students confusing 'absorbs more water' with 'drains faster'; the teacher gets an instant class-understanding summary and heatmap.
Scaffold: probe at the crucial point
AI prompts different students: 'Did the water stay, or pass through? Do the gaps between particles affect the flow?'
Reconstruct: upgrade understanding
From the evidence, students revise their judgment and use a diagram to explain particles, gaps and flow rate.
Co-create: put understanding together
In CocoNote a group investigates 'which plants suit which soils', forms a shared explanation and presents it.
After class: close the loop
CocoStudy pushes personalized review by performance; the portfolio flows back, and the teacher revisits weak points next lesson.
An upper-primary English lesson, 'Our lives in the future' — from one wild flight of imagination to a composition the student is genuinely happier with. Every step in between, the learning is seen, diagnosed, scaffolded and reconstructed.
Before class: set up the human–AI environment
In CocoClass the teacher loads the text and tasks, and uses CocoFlow to configure a 'future resident' chat agent and an essay-feedback agent.
Warm-up: let imagination enter the shared space
Students chat with the 'future resident' about future food, clothing and homes; curiosity is sparked and the topic enters the whole class.
Reading diagnosis: see who is stuck where
Students pick a title via multiple choice and sort information into 'diet/housing' via matching; the system returns accuracy instantly, so the teacher targets the hard sentences.
Draft: put understanding into writing
Students write their own 'blueprint of future life' and submit by photo or typing; dozens of handwritten drafts become cloud data in an instant.
Real-time feedback: scaffold at the crucial point
The essay-feedback agent returns a personalized diagnosis in seconds — corrections plus advanced-word and sentence/logic suggestions — and students revise.
Whole-class visibility: free the teacher from the red pen
On the teacher dashboard, submissions stream live alongside each student's AI dialogue; shared errors get a group nudge, individuals get face-to-face help.
Reflect & co-create: from 'learned' to 'learning how'
Students write self-assessments; the platform gathers the class's frequent words into a word cloud, and the teacher lifts 'future convenience' into 'technology and people'.
A grade-7 history activity lesson, 'Origins and heritage of traditional festivals' — three AI agents act like a 'scaffolding pyramid', lifting students from gathering facts all the way to designing a cultural event of their own.
Before class: build the scaffolding pyramid
Using CocoFlow the teacher configures three agents — a Festival Researcher, an Old-Painting Connoisseur and a Fair-Planning Assistant — and embeds them in the CocoClass lesson.
Active inquiry: let exploration enter class
Students choose Qingming / Dragon Boat / Mid-Autumn by interest and trace its origins with the Festival Researcher, turning passive listening into active questioning.
Socratic scaffold: questions, not answers
The Connoisseur doesn't explain directly but asks 'What is the figure doing with the willow branch?', guiding students to discover the custom themselves.
Deep understanding: write your own reading
Students become docents and write a caption for an old painting, upgrading 'watching the spectacle' into 'reading the meaning'.
Process visible: track the whole class
Dialogues are submitted in one tap; in CocoView the teacher reviews each group's conversation with the agents and tracks progress live.
Co-create: turn understanding into a plan
With a 6-step workflow, the Fair-Planning Assistant guides each group to complete a workable festival-fair proposal together.
An argumentative-writing lesson for middle and high school — AI doesn't write the essay for students; like a debate coach it pushes them to turn 'knowing the rules' into 'actually arguing well'.
Before class: line up three thinking coaches
Using CocoFlow the teacher configures three agents — an AI Debate, a Toulmin Argument Tutor and a Debate Sparring Partner.
Observe & model: see what good argument looks like
As judges, students watch two AI debaters spar pro and con, building a mental model of 'high-quality argument' in CocoNote.
Dual scaffold: build the argument step by step
The Toulmin Tutor guides students through claim, evidence, reasoning and rebuttal, then uses critical questions to make them self-check.
Retrieval under pressure: internalize
Students spar live with the AI partner, turning Toulmin skills from 'works on paper' into 'ready on the tongue'.
Thinking visible: turn the implicit into data
The system translates the whole debate into a strategy map, so students 'see' their own logical gaps for the first time.
Co-create review: from experience to reflection
In CocoNote students review the strategy map together and bank reusable argumentation experience.
SMART isn't only for students. In a district teacher workshop, teachers use the same method to go from 'I want to use AI' to 'I built my own teaching agent'.
Start: bring a real pain point
Nearly a hundred teachers put real teaching pain points from their subjects on the table, so the problem is seen first.
Design mentor: question an idea into a plan
CocoFlow's Design Mentor uses Socratic questioning to turn a fuzzy teaching idea into a clear agent design.
Build it: from user to creator
Without writing code, teachers express role, interaction rules and boundaries in natural language, building their own subject agents and interactive H5.
Test assistant: find problems before class
The Test Assistant role-plays a student to stress-test the agent and returns a coverage report on what still needs tightening.
Showcase & co-create: bank school resources
Teachers present and critique each other's agents, building a reusable, transferable school resource library.
A middle-school math practice lesson — AI makes 'correct-in-seconds' routine, but what really changes isn't speed: it's that every mistake gets seen, explained and put right.
Before class: set up the 'Basics Guard' grader
Using CocoFlow the teacher creates a math grading agent from a template and injects this lesson's answers and common error patterns.
Submit: surface the working
Students photograph their handwritten solutions; after OCR they enter CocoClass, so the process — not just the answer — is seen.
Correct in seconds: see where the class errs
The system checks against the answers instantly; in CocoView the teacher sees the error distribution and pins down the shared sticking point.
Error scaffold: not just wrong, but where
AI doesn't just mark a cross — it points out 'the sign flipped at step two' and gives a targeted next hint.
Revise & resubmit: turn errors into mastery
Students revise from the feedback and resubmit, turning 'got it wrong' into 'really get it' in a real-time loop.
After class: reinforce with similar items
CocoStudy pushes similar practice by error type; weak points flow back for review next lesson.
An IT lesson, 'drawing with 0s and 1s' — instead of memorizing concepts, students use plain language to 'build' a playable binary encoder, turning from tool users into creators.
Before class: make the abstract playable
In CocoFlow the teacher uses AI coding to generate a Minecraft-style binary-encoder H5 and embeds it in CocoClass.
Play then learn: concepts at the fingertips
Students click an 8×8 grid to toggle black and white and watch the matching 0/1 string update live; binary becomes something you can manipulate.
Layered exploration: a next step for everyone
Basic level just plays, intermediate edits parameters, challenge level sets its own task — AI gives just-right steps for each level.
Reconstruct: connect 'data' to 'image'
Students type 0/1 in reverse to generate an image, truly grasping how data represents information.
From user to creator
Students generate their own interactive mini-tools in natural language; the work is banked in CocoNote as a class resource library.
A middle-school physics lab on length measurement — the lab moves to the cloud, so every student can practice 'align the zero, read at eye level' over and over, error-free to try.
Before class: generate a 3D virtual bench
In CocoFlow the teacher uses AI coding to generate a 3D simulation H5 with a draggable ruler and embeds it in CocoClass.
Hands-on: make the procedure visible
Students drag a virtual ruler in the page to measure a block; the operation is shown live.
Error feedback: see the reading offset
If the ruler isn't aligned to zero or the line of sight isn't perpendicular, the system flags the error instantly; in CocoView the teacher tracks the class's command of the procedure.
Instant correction: prompt at the crucial point
AI prompts 'keep your line of sight perpendicular; align the zero mark first', handing the hard-to-supervise reading angle to the system.
Practice into procedure
Students practice 'aligning' and 'reading' over and over at zero cost, turning the rules from a mnemonic into muscle memory.
A primary-school art lesson — AI doesn't grade or nitpick; it focuses on catching the 'visual surprise' in each picture and gives children the warmest aesthetic response.
Before class: set up the 'Aesthetic Guide'
In CocoFlow the teacher creates an art-appreciation agent whose core logic is composition recognition plus highlight extraction.
Upload: let creativity be seen
Students photograph their artwork into CocoClass; every creation enters the shared view.
A warm response: catch the surprise
AI doesn't correct — it points out 'the clashing colors in your sky are bold', amplifying the child's spark.
Re-create: push the inspiration further
From the encouraging feedback, students adjust composition and color, turning one creation into one act of expression.
A high-school cross-subject project, 'campus carbon-footprint survey and redesign' — spanning geography, chemistry, math and IT, students collaborate on CocoNote to break one real big question into a workable plan.
Kickoff: put the real question on the board
On a CocoNote shared whiteboard a group asks 'how much carbon does our school emit a year?'; ideas, questions and roles are all visible at once.
Progress diagnosis: keep collaboration on track
CocoNote's monitoring agent tracks each group; in CocoView the teacher sees who is stuck on data and who on the plan.
Timely scaffold: probe at the crucial point
Planning and reflection agents step in: 'Is your reduction target backed by data? Is there a cheaper alternative?'
Reconstruct: rebuild the plan with evidence
From measured data, students revise their model and plan repeatedly, turning a vague idea into an evidence-based redesign.
Co-create: synthesize one proposal
Multi-subject roles converge on CocoNote into one shared carbon-redesign proposal and defend it.
After class: capability flows back
By each student's role in the project, CocoStudy pushes reinforcement tasks in the relevant subjects.
After a middle-school English unit test — CocoStudy turns a stack of wrong answers into a self-study loop of diagnose → path → coach → profile, so each student practises on their own weak points instead of the same worksheet.
Digitize the paper
Students photograph the unit test; OCR captures items and answers, and errors are structured into CocoStudy.
Error-cause diagnosis
The system separates careless, method and concept errors — telling 'tense confusion' apart from 'misreading the question' to find the real weak points.
Personalized path
From the profile it prioritizes review: shared weak grammar first, then similar and advanced practice per student.
Socratic coaching
When students ask, the agent gives hints to help them find what went wrong and why, rather than the answer.
Profile loop-back
This performance updates the long-term profile; weak-point shifts flow back to the teacher for a targeted next lesson.
Senior-year review — facing a mountain of problems, CocoStudy doesn't make students grind end to end; it diagnoses first, then lays out a precise review path by personal weak points so limited time goes where it matters most.
Diagnostic entry
Students take a baseline paper; OCR ingests it and the system breaks answers down by concept and item type.
Locate weak points
It separates shaky concepts, unpractised methods and calculation slips, producing a personal weak-point list and mastery profile.
Precise review path
Using the profile and exam-topic weighting, it orders review to tackle high-frequency, weak modules first.
Variant coaching
For each weak point it pushes similar and advanced variants; on follow-up the agent offers approach hints, not the answer.
Profile loop-back · next round
Each round updates the profile and re-tunes the path; teachers adjust their review focus from the class profile.
A primary-math unit — rather than 'redo what you got wrong', CocoStudy first works out whether a child was careless, hasn't mastered the method, or doesn't grasp the reasoning, then gives targeted practice.
Digitize the work
Students photograph their work; OCR reads each computation answer and files the errors.
Error-cause diagnosis
It separates 'copied the number wrong', 'forgot to carry' and 'doesn't grasp the reasoning' — telling carelessness from genuine gaps.
Targeted practice
Only the truly weak item types get similar practice — no 'drilling what they already know while missing what they don't'.
Profile builds up
Each round updates the mastery profile, so parents and teachers can see the progress trajectory.
A primary Chinese reading lesson — from one text to children who want to speak and write, CocoClass makes every child's idea visible, not just the few who raise a hand.
Before class: make the text interactive
The teacher embeds segmented reading, cloze and open questions in CocoClass, turning a static text into a class people can join.
Express: everyone gets a voice
Students post their views on a character to the wall; the whole class's ideas show live, giving quieter children a place to speak.
Diagnose: see the gaps in understanding
The system aggregates answers instantly; in CocoView the teacher sees who grasped the plot and who is stuck on words.
Co-create: gather the best lines
The platform pools the class's best expressions into a word cloud, and the teacher leads imitation and richer expression.
High-school chemistry has heavy problem sets and dense pitfalls — CocoStudy splits errors into concept / equation / calculation, so students know whether they erred on the chemistry or the math.
Digitize the paper
OCR import supports chemical equations and formulas (LaTeX); answers are structured into the system.
Layered error analysis
It distinguishes shaky concepts (e.g. redox), equation-balancing errors and calculation slips to locate the real sticking point.
Personalized path
From the profile it tackles high-frequency weak modules first, pushing similar and advanced variants.
Coached follow-up
On follow-up the agent gives approach hints (conservation, valence) to help students derive the answer.
Profile loop-back
Performance updates the profile; weak points flow back to the teacher for sharper review.
High-school English writing — instead of the teacher red-penning every essay, an essay-feedback agent gives each student instant diagnosis first, freeing the teacher for deeper coaching on ideas and expression.
Configure the feedback agent
In CocoFlow the teacher sets up an essay-feedback agent with three layers of criteria: grammar, structure and expression.
Submit & instant diagnosis
Students submit; the agent returns corrections plus sentence and logic suggestions, so they revise once on their own first.
Errors accumulate
CocoStudy logs frequent errors (tense, connectives, weak argumentation) into a personal writing profile.
Focused review
From the class profile the teacher addresses shared issues, saving one-on-one feedback for those who need it most.
A real cross-subject project at a Hong Kong middle school — 'survey and improve elderly residents' quality of life'. 132 grade-9 students work on CocoNote, going from a real social problem through survey, inquiry and brainstorming to solution design, ending with prototypes for real users.
Kickoff: put the social problem on the board
On a CocoNote open board, groups use topic, phase and task cards to break 'how to improve elderly residents' quality of life' into a workable project structure.
Survey & inquiry: phenomenon → problem
With resource, mind and camera cards, students gather community observations and interviews, turning scattered phenomena into researchable questions.
Brainstorm: ideas collide on the canvas
Groups share, link and negotiate ideas — dialoguing with AI to widen thinking — selecting feasible directions through idea conflict.
Progress & collaboration diagnosis
CocoNote's dashboard shows each group's progress, contribution and collaboration, so teachers step in and scaffold where groups get stuck.
Solution design: phenomenon-problem-product
From inquiry evidence students iterate solutions, producing prototypes for the target group — e.g. fall assistance, emotional companionship.
Co-create & present
Groups organize team results by linking cards and present their solution in an organized way using the present feature.
The real moat isn't front-end features — it's the shared intelligence core
The five products don't each grow their own AI. They share one Agentic-AI platform built around student portfolios, data assembly, resource knowledge and agent orchestration — the key to integrating fragmented data intelligently.
Two product spines run on one closed-loop engine, driven by the shared core.
Tap any node to see what it does — and when a teacher confirms.
Teacher uploads slides / courseware
Slide file + this lesson's goals
Break down concepts, find where interaction & observation can be generated
Structured courseware + suggested questions / observation tasks
One core, two kinds of growth
The core doesn't just connect data — it closes two growth loops: one helps students grow as learners, the other helps teachers grow as designers.
The teacher-side “see → refine” edge is on the way: feeding an agent's real classroom use and impact back into CocoFlow, so teachers iterate on the agents they build like they would on any creation.
Long-term student portfolio
Accumulates across lessons and products — the more it is used, the better it knows the student. A data network effect rivals cannot copy.
Data continuity
An error a student reveals in CocoStudy is visible to the teacher in the very next CocoClass lesson.
AI proposes · human confirms
On key judgments AI suggests first; the teacher keeps the final say — trustworthy enough for a real classroom.
What SMART brings to teachers, students and schools
Less guessing, sharper judgment
- ✓See how students think, faster
- ✓Spot shared misconceptions earlier
- ✓Differentiate support more easily
- ✓Organize higher-quality discussion
- ✓Keep teacher leadership and final say
Not just finishing tasks — really growing
- ✓More willing to express ideas
- ✓Clearer about what they don't understand
- ✓Scaffolding fit to their state
- ✓Deeper understanding via revision and transfer
- ✓Learning to express, listen and co-create
From buying tools to upgrading the environment
- ✓A unified AI-native learning environment
- ✓Connecting class, study, projects and workflows
- ✓Sustainable teaching data and learning assets
- ✓Support for teachers' AI literacy
- ✓A systemic framework for the smart classroom
Resource-generation tool vs. learning-loop system
General generative tools are strong at producing materials but weak at managing process. The Coco system is process-oriented — built to run the classroom and learning loop.
| Capability | General generative tool | Coco system |
|---|---|---|
| Resource generation | Strong | Available |
| Task-driven | Weak | Core |
| Feedback & intervention | Weak | Strong |
| Deep diagnosis | None | Core |
Coco solves 'did the student actually learn'.
What SMART is not
Interaction alone
A busy classroom with many clicks, but thinking left unseen, is not SMART.
Data alone
Plenty of dashboards, but the teacher still doesn't know what to do next, is not SMART.
Recommendation alone
Different problems pushed, but no scaffolding and no help understanding, is not SMART.
Reformatting alone
Turning text into charts, cards or podcasts without students re-understanding is not SMART.
Grouping alone
Sitting together without a shared goal and a shared outcome is not SMART.
AI literacy: dedicated tracks for students and teachers
Environment and people are the two halves of AI-native learning — complementary and mutually essential: the SMART product suite builds the AI-native learning environment, while students' and teachers' AI literacy is the human side. So we also build structured AI-literacy learning and assessment for schools, each unfolding into its own dedicated page.
The courses and assessments aren't designed in a vacuum — they are built on authoritative frameworks: aligned with China's MOE student-facing AI General-Education Guidelines for Primary & Secondary Schools (2025) and the teacher-facing Digital Literacy of Teachers standard (2022, covering AI-related competencies), UNESCO's AI Competency Frameworks for students and teachers (2024), and the OECD–European Commission Empowering Learners for the Age of AI framework (2025) — then iterated with frontline classrooms and learning-sciences research.
Student AI literacy
Helping students understand, use, evaluate and govern AI more responsibly — not just use it, but judge it.
- ▸AI general-education course — A structured AI-literacy course and learning experience for students.
- ▸CocoPi — AI hardware with local-model support, ready to teach out of the box, bridging to AI+STEM practice.
- ▸Student AI-literacy assessment In development — An evidence-centred literacy assessment instrument.
Teacher AI literacy
Helping teachers grow from 'AI tool users' to 'AI learning-system designers'.
- ▸Teacher AI-literacy course — An AI-literacy and classroom-application course for frontline teachers.
- ▸Teachers as agent creators — Teacher PD and creation practice already run in Pingshan, Changzhen and beyond.
- ▸Teacher AI-literacy assessment In development — A literacy and AI-TPACK assessment for teachers.
'Teachers as agent creators' is not a slogan — these interactive subject sites are real artifacts co-created in 'AI for teachers' training for subject curriculum leaders. Completely free and ready to use; Primary Science is live now, with more subjects on the way.
AI-empowered Primary Science · interactive learning site
Built for primary-science teachers, spanning physical, life and earth & space science inquiry.
Open courseware ↗More subjects coming
Chinese · Math · English · IT — rolling out alongside subject-sharing sessions.
Change that's already happening in real schools
These aren't hypotheticals — they are teacher-empowerment programmes and classroom practices already running across Shenzhen districts and schools (Pingshan, Changzhen, Futian, Bao'an), plus classroom cases documented by CocoRobo Research.
District-level AI agent-creation workshop
~100 teachers joined a two-day workshop on subject-specific agent development, workflow design and AI programming — and independently designed and presented their own agents.
A full theory–practice–creation–presentation cycle
Seed teachers moved through the full cycle, creating pedagogical agents, workflows and H5 learning products into a reusable school resource base.
AI-assisted biology teaching cases
Lessons supported by agents, interactive pages, formative diagnosis and personalized suggestions — with students iterating on their own designs.
Teach–learn–assess, deeply fused (English)
In a 'future living' English lesson, CocoClass's board, multiple-choice and matching items and submissions pool whole-flow data while CocoFlow's Future agent answers as a 'learning partner' — a teacher–student–AI triad turning pre-reading, drafting, feedback and revision into a live teach–learn–assess loop.
From 'blind box' to 'live broadcast'
A CocoFlow-built essay-review agent is embedded in the lesson so students iterate write–review–revise on the spot; CocoClass shows every student's submission and AI dialogue live on the teacher's screen, turning class progress into a broadcast so the teacher lands beside whoever needs help.
Toulmin scaffolds + an AI debate room
Grounded in cognitive science, CocoFlow builds a 'Toulmin argumentation tutor' (a build-then-critique double scaffold) and an 'AI debate' where two AI debaters spar across rounds while students judge — moving from memorizing rules to actively modelling sound reasoning.
AI lesson review: a science lesson made legible
The same teacher taught one topic twice; CocoView reconstructed each lesson with the activity spectrum and problem-chain tree — the second time, focused inquiry stretched from fragmented bursts to about 15 minutes and forced interruptions dropped from two to one, shifting interaction from alternating to modular.
Three agents, three layers of deep learning
In a 'roots of traditional festivals' activity lesson, a CocoFlow workflow uses a form card to let students pick a festival and a control card to route them to Qingming / Dragon-Boat / Mid-Autumn expert agents that unfold by inquiry rather than spoon-feeding; in 8 minutes students build and upload a festival timeline — three dialogues, three layers of deep learning.
A 132-student community-elderly PBL on one canvas
At a Hong Kong middle school — 'survey and improve elderly residents' quality of life' — 132 grade-9 students use topic/phase/task and resource/mind/camera cards on a CocoNote canvas to move from a real social problem through survey, inquiry and brainstorming to solution design, ending with prototypes for real people.
Research-informed, practice-tested
The SMART principles come from a combined reading of real classroom problems, learning-sciences research and product practice. CocoRobo works closely with frontline schools, teachers and university researchers on AI-supported interaction, self-paced study, collaborative learning and teacher AI-workflow design.
CocoRobo Research · research & practice platform
For scholars, long-term university partners and deep education researchers — research questions, methodology, cases, publications and collaboration paths for AI-native learning systems, all in one place.
- Six pathways from research to school practice
- Products as research platforms
- Five long-term research questions
- Design-Based Research methodology
- Publications archive
- How collaboration starts
Let's build a smart classroom that truly has quality
If you're thinking about how to bring AI into teaching in a way that supports teachers and develops students, CocoRobo would love to explore it with you.