On Ground Labs

Focus Areas

The Hypothesis

AI research today is too abstract, too distant from everyday realities.

We believe intelligence should emerge from field work, real people, and persistent context. On Ground Labs exists to build systems that stay close to the world they serve—human-first, rigorously empirical, unapologetically practical.

Exploring

Small Language Models

Deployable intelligence that runs at the edge or in low-bandwidth classrooms across India.

Field Intelligence Loops

Embedding researchers with schools and labs to gather the signals academic papers miss.

Mentor Tooling

Lightweight workflows that help research mentors co-author work with students in real time.

∞

Papers Open Forever

“Intelligence should stay grounded in people, purpose, and the messy, vibrant world we share.”

— The Philosophy

Launch Blueprint

On Ground Labs is charting a launch for 2026, grounded in field work with students, research partners, and public institutions. This is Tanay Pratap's initiative and, while in exploration, we're intentionally not seeking external funding yet.

The blueprint for the coming months is evolving through a set of academic and partnership experiments:

  • Q4 2025: Train small language models from scratch, then LoRA-supervised finetune, and finally RL finetune them to understand the entire stack from first principles.
  • Partnerships: Explore collaborations with budget schools (≤ ₹20K monthly fee) to ground pilots in classrooms. If you can connect such institutions, please reach out.
  • Applied research: Benchmark and finetune current SLMs like Qwen and Gemma for Socratic learning of mathematics and Python, and probe agentic development and deployment patterns.

The plan is to consolidate findings through Q4 2025 and early Q1 2026, then publish the work and deploy solutions with partner institutions. If you want to contribute, write to tanay.mit@gmail.com.

Students collaborating around a table with laptops, viewed from above.
The Vision

Publish research that is instantly actionable for teachers.

Partner with Indian universities to co-run open labs.

Develop public datasets rooted in lived community knowledge.

Equip students with the tools to lead their own inquiries.