Inspiration

Lumi started with a simple question: what happens when a learner needs more than an answer?

As a parent supporting three children across different ages, learning needs and languages, and as a university student myself, I have seen how easily learners can become stuck when educational material cannot adapt to them.

So naturally, I wanted to explore an AI learning experience that could sit alongside the learner: reading what they are learning, asking thoughtful questions, identifying the point of confusion and helping them work towards understanding without simply completing the work for them.

What it does

Lumi is a trusted, multimodal AI learning companion designed to help learners understand schoolwork, course materials and difficult concepts across subjects, languages and educational stages.

Learners can bring Lumi something they are studying, explain where they feel stuck and choose how they would like support:

Teach me Give me a hint Check my work Help me revise.

Rather than immediately generating an answer, Lumi asks diagnostic questions to discover what the learner already understands. She uses the learner’s responses to identify misconceptions, missing foundations or alternative ways of approaching the concept.

How we built it

Codex acted as my engineering partner throughout the build. It helped translate the product vision into a working application, scaffold the experience, implement the tutoring flow, structure the interface, connect the different learning surfaces, build lesson-history functionality and continuously diagnose and improve the application.

GPT‑5.6 powers Lumi’s central teaching intelligence. It is used to:

Interpret learning materials and learner questions; Assess the learner’s current understanding; Generate purposeful diagnostic questions; Identify misconceptions; Select an appropriate teaching strategy; Adapt explanations to the learner’s stage and language; Produce structured content for the Lumi Lab visual whiteboard; Check understanding through reflection and learner explanation; Generate lesson recaps and suggested next steps.

The tutoring experience follows a deliberate learning loop:

Understand → Question → Diagnose → Adapt → Explain → Check → Reflect

Challenges we ran into

I had issues trying to integrate LemonSlide to animate the live chats feature, but I'll eventually get it right and that will be an updated feature in Lumi V2

Accomplishments that we're proud of

I loved this project so much as it was close to my heart to begin with and experiencing every moment of pushing forward, making progress and overcoming technical challenges to bring to life and be able to submit my entry was exactly what I'm most proud of.

What we learned

That I'm obsessed building!

What's next for Lumi

Watch the space because there is a lot more coming soon, we can't just stop here now can we!

Built With

  • 2
  • api
  • appwrite
  • codex
  • file
  • github
  • gpt
  • gpt-4o
  • gpt-5.6
  • image
  • mini
  • next.js
  • openai
  • react
  • realtime
  • responses
  • search
  • sites
  • stores
  • transcribe
  • tts
  • typescript
  • vector
  • webrtc
Share this project:

Updates