About the Project — Intueri

Inspiration

Creating clear STEM animations takes hours. We wanted a way to turn plain-English explanations (e.g., “derive $\sin(x)$”) into clean visuals in minutes. That became *Intueri *: prompt → animation, optimized for accuracy and speed.

What it does

Intueri converts natural-language prompts into Manim animations and returns a rendered MP4.

1. User enters a prompt (e.g., “integration of $\sin(x)$”).
2. An LLM generates Manim (Python) scene code.
3. A queue triggers a renderer that produces a video with LaTeX, graphs, and geometry.
4. The user downloads/plays the MP4.

How we built it

• Frontend: Next.js/React (status polling, job queue UI).
• Backend: FastAPI (Python) exposing /api for job submit/status.
• Queue: Redis + Celery for async job processing.
• Renderer: Manim in Docker; outputs 1080p@60fps MP4s.
• Database: MongoDB (prompts, scripts, metadata).
• Infra: Docker Compose; deployable on GCP.

Challenges we faced

• Sharing a /shared volume between worker and renderer containers.
• Making LLM-generated Manim code both syntactically valid and mathematically correct.
• CORS/config in local dev; real-time polling for render status.
• Debugging Manim scene errors and stabilizing resource usage.

Accomplishments we’re proud of

• Successful renders for multiple math prompts (e.g., $\int \sin(x)\,dx$).
• Clean, modular UI with live status.
• Solid containerized architecture that’s easy to scale.
• AI text-to-speech voiceovers and subtitles.

What we learned

• Manim scene composition (axes, LaTeX, transforms).
• Productionizing LLM output safely into executable code paths.
• Container orchestration with Docker Compose and Redis queues.
• Fast iteration under hackathon constraints.

What’s next

• Multi-language prompts and captions.
• Public “Prompt-to-Lecture” library of shareable scenes.
• Render scaling via cloud workers.

Built With

• docker
• gcp
• mongodb
• next.js
• python
• redis
• manim
• fastapi

Built With

• docker
• gcp
• mongodb
• next.js
• python
• redis