IlmuEducator

Inspiration

Teachers spend hours preparing lessons before they can teach. They need to interpret curriculum requirements, plan objectives, create activities, prepare worksheets, anticipate misconceptions, and adapt material for different learners.

IlmuEducator was built to reduce that preparation burden without replacing teacher judgment. The goal is a teaching workspace that turns a lesson idea or source document into a complete, reviewable learning experience.

What It Does

IlmuEducator turns a teacher's country, grade, subject, topic, language, learner profile, objectives, and optional source material into a classroom-ready lesson package.

It creates interactive quizzes, matching games, printable worksheets, answer keys, flashcards, study notes, mock assessment practice, teaching guidance, a learning map, and a lesson quality review.

Material Intelligence turns uploaded or linked material into a source-grounded summary, topics, key terms, evidence, and learner explain-goals. When source text is limited, IlmuEducator labels that limitation instead of inventing detail.

IlmuEducator is local-first. Teachers can save lessons, use Teach Mode, and record reflections on their own device without student accounts or a cloud database.

How We Built It

IlmuEducator uses a React and Vite frontend with custom CSS, plus a Node.js and Express backend.

The teaching workflow has five focused stages. Curriculum Intelligence identifies objectives, concepts, misconceptions, prerequisites, and curriculum context. Experience Designer selects an activity strategy. Content Generator produces interactive and printable learning resources. Teacher Assistant creates classroom guidance and differentiation ideas. Lesson Evaluator checks learning coverage, engagement, difficulty balance, and Bloom's taxonomy.

The system uses structured schemas, validation, retry handling, source safety controls, and transparent standards boundaries. It distinguishes reviewed curriculum outcome matches from framework-only guidance instead of claiming universal official alignment.

Codex and GPT-5.6 were used as engineering partners for product decisions, architecture, implementation, debugging, and verification. A key decision was to build reusable browser game engines once instead of generating fragile game HTML for every lesson. That made activities faster, portable, and more reliable.

During Build Week, IlmuEducator was meaningfully extended with local-first Teach Mode and lesson history, worldwide curriculum pathways, source-registered standards boundaries, offline language packs, and source-grounded Material Intelligence.

Challenges We Faced

Reliable educational output needs more than fluent text. Activities require consistent structure, clear answers, suitable difficulty, and usable classroom formatting. We addressed this with structured output schemas, validation, retries, and teacher review boundaries.

Curriculum accuracy was another challenge. IlmuEducator only presents an exact curriculum match where a reviewed outcome record exists. Otherwise it clearly asks the teacher to review the official framework.

Language support also requires honesty. IlmuEducator provides verified offline language packs where available and makes fallback behavior visible rather than claiming identical coverage for every language.

What We Learned

Educational AI needs transparency, teacher control, and clear limits. Reusable systems, validated outputs, and explicit trust boundaries are more useful than one-off generated content.

AI is most valuable when it supports the teacher's expertise. IlmuEducator helps teachers prepare faster while teachers remain responsible for reviewing content, choosing what fits their class, and guiding learning.

Built With

• codex
• express.js
• gpt-5.6
• javascript
• node.js
• openai
• react
• vercel
• vite
• zod