Doceo - AI STEM Tutor

Doceo turns any STEM problem into an interactive, step-by-step lesson with clear derivations, adaptive pacing, and realistic voice narration so students can learn faster and retain more.

Comment

About the Project

Inspiration

Most AI tutors either dump a final answer or give long text that’s hard to follow. We wanted something that feels like a real guided lesson: clear step flow, visual derivation, and the ability to ask questions in context while learning.

What It Does

Doceo takes a STEM prompt (typed or image-based), generates a structured step-by-step lesson, and plays it as an interactive walkthrough.

Users can:

  • Follow derivations live, step by step
  • Pause and replay explanations
  • Ask quick follow-up questions during the lesson
  • Generate exam-cram plans from uploaded materials
  • Revisit previous sessions from history

How We Built It

We built Doceo as a full-stack application.

Frontend

  • Next.js
  • React
  • TypeScript
  • Framer Motion
  • KaTeX

Backend

  • FastAPI
  • Server-Sent Events (SSE) streaming
  • Pydantic

AI

  • Gemini for lesson generation
  • Contextual chat
  • Voice narration (TTS)

Core Systems

  • Timeline player
  • Step renderer
  • In-lesson interruption flow
  • Exam cram plan generator
  • Session history

Challenges We Ran Into

  • Making lessons feel interactive and polished instead of robotic
  • Keeping layout deterministic and readable over long derivations
  • Preventing audio/visual drift in chunked narration playback
  • Handling model variance so outputs stay structured and useful
  • Merging major parallel branches without regressing UX

Accomplishments We’re Proud Of

  • End-to-end interactive lesson flow from prompt/image to guided derivation
  • In-lesson interruption (“quick ask”) that preserves lesson context
  • Gemini-based voice narration integrated with lesson playback
  • Exam Cram mode and History mode integrated into product navigation
  • Material-grounded cram fallback behavior (not canned default content)

What We Learned

  • Learning UX quality is part of correctness, not just visual polish
  • Deterministic state and replay behavior matter as much as model quality
  • Audio and visual sync requires strict orchestration boundaries
  • AI systems need strong normalization and fallback design to feel reliable

What’s Next for Doceo – AI STEM Tutor

  • Further improve long-session playback reliability and sync
  • Add smarter revision workflows in history (search, tagging, spaced review)
  • Expand adaptive practice generation from user performance patterns
  • Continue mobile-first polish for the full interactive lesson experience

Built With

Share this project:

Updates