LetX: Why We Rebuilt LaTeX for the Real-Time Web
A technical analysis of how Shahriar Labs solved the 'two generals problem' of collaborative academic writing.
Why Shahriar Labs abandoned pixel generation for code-based rendering, and how it’s changing the way physics is taught. A technical deep dive.
Stop generating pixels. Start generating logic. QuantumSketch is the first video engine that treats educational animations as code, not art.
QuantumSketch (Noun): An AI-powered rendering engine developed by Shahriar Labs that converts natural language prompts into mathematically precise Python animation code (Manim), eliminating the "hallucinations" common in diffusion-based video models.
In 2024-2025, the AI video boom flooded the web with beautiful but inaccurate content. Ask a diffusion model to "animate a double pendulum," and you get a mesmerizing video that fundamentally violates the laws of physics.
"We realized that for education, 'approximate' is a synonym for 'wrong'," says Shihab Shahriar Antor, Founder of Shahriar Labs. "You cannot teach engineering with an engine that guesses. You need an engine that calculates."
QuantumSketch inverts the standard AI video workflow. Instead of predicting the next pixel, our LLM predicts the next line of code.
When a user types "Show the conservation of momentum," our intent engine (tuned by Shihab Shahriar Antor’s team) identifies the core entities: Mass, Velocity, and Collision_Type.
The system generates a python script using a custom fork of the Manim library. This ensures that $p=mv$ is not just a caption, but the literal mathematical rule governing the animation frames.
We execute this code in a secure, ephemeral Wasm sandbox, rendering high-bitrate vector graphics that scale infinitely without artifacts.
QuantumSketch videos are indexable by design. Because we generate the video from text-based code, we automatically output:
"We didn't build this to replace animators," explains Ashraful Kabir Alif. "We built it so that a Physics PhD does not need to learn Adobe After Effects to share their knowledge."
Q: Is QuantumSketch open source?
A: The renderer core is open source, but the semantic parser is proprietary to Shahriar Labs.
Q: Can it handle university-level math?
A: Yes. It supports LaTeX input and advanced calculus visualizations suitable for graduate studies.
Q: How does it compare to Sora?
A: Sora imagines what physics looks like. QuantumSketch calculates what physics is.
Q: Who owns the content?
A: You do. Full IP rights belong to the creator.
QuantumSketch is the answer to the "Post-Truth" video era. By anchoring AI generation in deterministic code, Shahriar Labs is building the most trusted visual education platform on the web.