Project Rainbow · Demo 6
Grandpa Tam: Then and Now
A complete KS2 English lesson, performed, coached, filmed and assessed by a team of autonomous AI agents inside Minecraft. No actors, no editing, no script read aloud: every line you hear is the agents thinking live.
One continuous, unedited take (2 min 31 s). The camera work, the conversation and the closing teacher report were all produced by the agents.
The lesson
A young learner, Pip, meets Grandpa Tam, the village's oldest fisherman, at the harbour. The task: find out how life was then and how it is now, ask at least three questions, and thank him in your own way.
Grandpa Tam: "I felt very sad, Pip. My first boat sank in a storm. Good friends helped me build a new one."
Along the way the learner practises asking open questions, listening, comparing past and present, asking about feelings, and expressing gratitude, then leaves Grandpa Tam a thank-you sign and writes in a reflective journal.
Built for the Hong Kong curriculum
English Language
Module "Changes", unit "Now and then: a changing world, technology". Interviewing a person from the past with simple present and simple past.
Empathy & gratitude
Directly serves the EDB objectives of comparing then and now and developing empathy and gratitude.
Following EDB guidance
Matches the EDB examples of AI in learning and teaching: an AI avatar role-play plus an AI writing companion that hints rather than gives answers.
The same engine runs a growing pack of EDB-aligned scenarios across subjects and key stages.
A team of AI agents, running the whole lesson
Project Rainbow is an agentic workflow: specialised agents work together, each with a clear role and clear boundaries, and the lesson runs end to end without a human operator.
🎭 Character agents
Grandpa Tam lives in the world, remembers his life story and responds to how he is asked, opening up when the learner shows care.
🧒 Learner agents
Synthetic students with distinct personalities rehearse the lesson, so every scenario is tested before it reaches a classroom.
🧭 Coach agent
A private, kind coach who only whispers to the learner, nudges rather than tells, and gives a short debrief at the end.
🎬 Director agent
Stages the scene, moves the cast, keeps the lesson on track, and directs the camera like a film crew: wides, two-shots, over-the-shoulder, close-ups.
👁️ Observer agent
Watches what the learner actually says and does, and turns it into evidence for the teacher.
📊 Teacher insight
Each session ends with a clear, evidence-linked summary: what went well, what to try next, and where help was needed.
The learning loop
When a learner gets stuck, the coach steps in privately and the learner tries again, so every child gets personal guidance in the moment, and the teacher sees exactly where it was needed.
Inside the film
Technology highlights
Fully autonomous production
Set building, casting, staging, dialogue, coaching, camera direction, recording and reporting all run from a single command.
Live generative dialogue, safely bounded
Characters speak freely in their own words, while the lesson structure, age-appropriate content and assessment stay under the platform's control.
AI film direction
A virtual director reads the scene and calls the shots in real time, avoiding walls and bystanders like a camera operator would.
Evidence-based insight
Every judgement in the teacher summary is linked to what the learner actually said, never to a guess about who they are.
Runs in a world children love
A real Minecraft server, real characters, a real place to explore, with the learning designed in.
Scales across a class
Many learners, each with their own personality and path, interviewed in turn with a summary for the whole class.
Tech spec at a glance
World
Minecraft Java Edition server, containerised
Intelligence
Large language models, provider-agnostic
Architecture
Multi-agent orchestration with a central director
Agents
Character · learner · coach · director · observer
Avatars
Animated 3D characters in a live web view
Camera
AI-directed virtual cinematography, HD capture
Outputs
Lesson film, live captions, teacher insight report
Deployment
Runs on a standard PC; cloud-ready
Responsible by design
Coaching is private and never judges the child, only the behaviour. Hints guide rather than give answers. Language accuracy is kept separate from social skills, so emerging English speakers are assessed fairly. The learners in this film are AI personas used to rehearse and test the lesson; classroom pilots with real learners come next.