PadhAI

PadhAI: From prompt to narrated and animated lectures, instantly.

Comment

Inspiration

  • “PadhAI” (padhai = “study” in Hindi) was born from a simple idea: turn any topic into a clear, animated mini‑lecture—instantly.
  • Teachers and learners spend too much time storyboarding, animating, and narrating; we wanted studio‑quality explanations without production overhead.

What it does

  • Converts a prompt into a time‑aligned lecture video: Manim animations + synchronized narration.
  • Uses a shot‑based timeline (scenes → shots with start times) to guarantee narration/visual sync.
  • User friendly UI with chat history, instant replays, and browser TTS.

How we built it

  • Frontend: React + Vite (VideoPlayer, ChatPanel, Sidebar), Web Speech API for TTS scheduling by shot start times.
  • Backend: Node.js + Express; calls an OSS LLM via Groq; assembles Manim code; executes Manim; serves generated MP4s.
  • Prompting: “Maestro” system prompt returns JSON with manim_header, scenes[shots], and manim_footer.
  • Reliability: Code sanitization (quote fixes), camera‑animation shims for Manim version differences, Windows‑safe cleanup.

Challenges we ran into

  • Manim compatibility: camera.animate not available on some builds; added safe fallbacks and timing preservation.
  • JSON → Python hygiene: over‑escaped quotes breaking strings; added robust unescaping and indentation normalization.
  • Windows file locks (EBUSY) during cleanup of partial media; added resilient deletion and retries.
  • Timeline trust: ensuring animation durations meet narration length; enforced waits/padding.

Accomplishments that we’re proud of

  • Shot‑based timeline that makes narration and visuals deterministically sync.
  • Clean, accessible UX that feels familiar and fast.
  • Fully local render pipeline (Manim + browser TTS) using free tiers.
  • Defensive backend: sanitization, fallbacks, and portability fixes.

What we learned

  • LLM‑to‑code is powerful but fragile—small escaping/indent errors can break renders.
  • Cross‑version portability in Manim requires careful camera handling.
  • Windows dev ergonomics (file locking, CRLF) need explicit handling.
  • Clear contracts (timeline schema) make complex AI outputs reliably usable.

What’s next for PadhAI

-Fine tune OpenAI GPT OSS with manim codes and synchronized narrations

  • Real TTS streaming and audio–video mux (ffmpeg) for shareable, packaged lectures.
  • Template library for common pedagogy patterns (graphs, processes, proofs).
  • Sandbox execution (Docker/containers) and security hardening.
  • Persistence (saved prompts, edits, versions) and collaboration.
  • CI for lint/tests and a one‑click cloud deploy path.

Built With

Share this project:

Updates