[Proposal] Extending PaperTalker for Generative Education (Zulense Collaboration)

[ProgramerSalar]

Hi @showlab

First off, huge kudos on PaperTalker (arXiv:2510.05096). I've been digging into the code here and the approach to "Visual Anchoring" and the "Cursor Builder" is brilliant. It solves a major problem in how we ground AI outputs in structured content.

I am Manish Kumar, Founder of Zulense (Pre-Seed Funded). We are building India's first Vertical GenAI Engine for education, and I see a massive opportunity to apply your research to a real-world problem.

The Research Problem
Current SOTA models (Veo, Wan-ai) are built for "Cinema." They excel at lighting but fail at "Pedagogy."

If I ask Veo to generate "A teacher writing letter 'A' on a blackboard," it hallucinates the text and stroke order. It generates pixels, not lessons.

Our Approach: The "Teacher Brain"
We are building a proprietary architecture that handles fine-grained control (Text Rendering, Stroke Order) separately from the background imagination (using large diffusion models). We are moving beyond simple video generation to instructional generation.

The Proposal
I’d like to invite you to join us as a Founding AI Scientist (Remote/Advisory) to guide this architecture.

You would act as the "Architect" while our team handles the execution:

The Math: I have onboarded two excellent IIT students (Top 1% in India) to handle the mathematical heavy lifting and paper implementation.

The Code: I am a Senior AI Engineer myself (leading the architecture), supported by two AI Engineers focused purely on execution, training pipelines, and MLOps.

The Offer:

Competitive Retainer: ($2.5k/mo range)

Founding Equity

Compute & Data: Access to our unique handwriting dataset & compute resources.

Your research shouldn't just stay in a PDF (or this repo); let's use it to teach 10 Million students how to read.

Are you open to a 10-min chat this week?

Best,

Manish Kumar
Founder, Zulense
zulense.com