ABOUT SPELLTAG

A platform for exploring what generative AI can add to physical educational toys.

SpellTag begins with a familiar learning action: arranging alphabet cards to form a word. A Wi-Fi-connected reader carries that physical input into a browser platform, where GPT Image produces a visual interpretation for discussion and feedback.

WHY GENERATIVE AI

A response that is adaptable—and worth questioning

Fixed flashcards are predictable and dependable, but they cannot respond to every word a child chooses. Generative AI can produce a new illustration on demand.

That flexibility introduces uncertainty. SpellTag makes the uncertainty useful by asking learners to compare the word and image, notice mistakes, and explain how an instruction could improve.

CURRENT ARCHITECTURE

The existing physical prototype, connected over Wi-Fi

The competition build keeps the ESP32-S3, PN532 reader and NFC alphabet cards while improving the software platform around them.

NFC alphabet cardsHands-on word construction
ESP32-S3 readerWi-Fi station mode
Cloud platformRealtime plus polling fallback
Learning experienceGPT Image, gallery, missions, feedback

SAFEGUARDS AND LIMITATIONS

Responsible use requires controls, transparency, and an adult in the loop.

SpellTag uses safeguards; it does not promise perfect safety or accuracy.

Controlled production provider

New image generation is locked to GPT Image. Provider switching is not exposed to users or administrators.

Child-oriented instructions

The server builds a constrained illustration request and avoids presenting generated output as a factual answer.

Visible feedback loop

Learners can reject a result and identify why it was inaccurate, generic, or stylistically wrong.

Supervised use

SpellTag is intended for guided home, classroom, and learning-support activities—not unsupervised advice.

Known limitations

  • AI images can misunderstand ambiguous or unusual words.
  • A network and available generation service are required for new images.
  • Prompt and provider controls reduce risk but cannot guarantee appropriate output.
  • Long-term learning outcomes have not yet been established.

DATA AND PRIVACY

Collect only what the experience needs

Guest use and account use are separated. Generated words, images, feedback, and progress may be stored to provide galleries and learning features.

Device credentials, Wi-Fi passwords, service keys, raw command payloads, and infrastructure identifiers must never appear in public pages, screenshots, or submission materials. Production access is being hardened as part of this upgrade.

PROJECT STATUS

Where the platform stands today

Implemented end-to-end prototype

The physical cards, reader, Wi-Fi event path, browser platform, generation flow, gallery, missions, and feedback system have been implemented. A final end-to-end competition rehearsal is still required before submission.

A+ final-year project

The project received an A+ academic result before its current production and competition upgrade.

Validation continues

Public usability work and educator-guided pilot preparation continue. Long-term educational benefit is not claimed yet.

FROM COURSEWORK TO COMPETITIVE ENTRY

The next version focuses on public trust and production quality.

Following the Open SPACE demonstration, the team is strengthening security, cost control, input flexibility, dictionary quality, admin workflows, accessibility, pilot preparation, and the public product story. The physical hardware remains unchanged.

Open the current platform