Agentic AI

Mesmerising Playground for School Learners

Explore how AI agents reflect, plan, act, and learn — with cute visuals, gentle 3D magic, and colourful steps you can follow.

Human-in-the-loop
Orchestration
Safety
Time & Cost
Multi‑Agent
✨ Play with the 3D view!
Cognitive Loop — colourful steps
Reflect → Plan → Act → Critique → Update (with teacher checkpoints)
1. Reflect
  • Think about goals, constraints, and safety signals.
  • Rubric: truthfulness, harmlessness, usefulness.
  • Checkpoint: ask a teacher/mentor for approval.
2. Plan
  • Draft a mini‑plan with approval gates.
  • Break down into tiny sub‑tasks.
  • Mark risky parts for extra review.
3. Act
  • Execute small, reversible actions with logging.
  • Prefer safe changes that are easy to undo.
4. Critique
  • Self‑review with rubrics and tests.
  • Ask: Safe? True? On‑policy?
5. Update
  • Revise plan, store learnings, and iterate.
  • Improve your next run!
Agent Orchestration
Manager/Worker roles with guards & recovery
Long‑term Memory
Personas • Policies • Project State
Persona: Helpful Tutor
Kind, simple words, emojis ok
Policy: School‑Safe
No personal data; no unsafe code
Project State
Lesson #3 in progress
Retriever
Keeps facts fresh
Memories
Stores learnings & mistakes
Notebook
Cites sources when needed
Observation & context tracking let agents remember preferences over time.
Evaluation Metrics for Autonomy
Success • Safety • Cost • Time
60%
Safety Snapshot
Guardrails & rollback friendly
Higher autonomy can speed things up, but stay inside policy bounds. Use checkpoints!
Run the Agent
Action execution with rollback, guardrails, and logging
System ready. Policies loaded. 🛡️
Cute Gallery & Inspiration
Cartoons and AI‑style imagery to spark curiosity
Tip: Make mascots! Planner Panda, Critic Cat, Memory Mouse. 🐼🐱🐭
Core Concepts — quick map
Human‑in‑the‑loop alignment
Adults/teachers approve key steps before continuing.
Evaluation metrics for autonomy
Success, Safety, Cost, Time — balance these!
Self‑critique mechanisms
Reflect, diagnose, revise. Ask: what did we miss?
Agent orchestration frameworks
Pipelines, guards, and recovery flows to stay safe.
Observation & context tracking
Remember history across sessions.
Hierarchical architectures
Manager/worker agents with specialised roles.
Autonomous goal setting (bounded)
Set small goals that follow policy.
Self‑reflection & step reasoning
Explain steps clearly and transparently.
Multi‑agent collaboration & roles
Planner, Executor, Critic — teamwork!
Dynamic planning & replanning
Update plans when facts change.
Long‑term memory
Store personas, policies, and project state.
Action execution with rollback
Always log & keep an undo button.
Feedback & adaptation
Improve across deployments with lessons learned.
Ready to try your own mini‑agent?

Pick a tiny task (like summarising today's lesson). Follow the colourful loop: Reflect → Plan → Act → Critique → Update. Ask a teacher to approve checkpoints.