The Beginning
It started at the India AI Impact Summit 2026 in New Delhi. That's where I met Dawar (who would go on to found the lab) and Dhruv - and we realized we shared the same conviction: that the next leap in AI won't come from scaling alone, but from understanding and compressing what models actually learn.
What began as conversations at the summit turned into Erdős AI Lab - named after Paul Erdős, the mathematician who believed that knowledge should be open, collaborative, and relentlessly pursued. We exist at the intersection of rigorous mathematics and unbounded curiosity - a collective of young researchers devoted to advancing the frontier of AI. We believe the next decade will not be won by scale alone, but by ideas. Sharp, original, uncompromising ideas. That is what Erdős is built for.
What We're Building
Our research is focused on three interconnected areas:
Knowledge Distillation
We're developing pipelines that compress large teacher models into efficient student models without meaningful performance loss. This isn't just about making models smaller - it's about understanding what knowledge transfers and what doesn't, and engineering better compression strategies based on that understanding.
Representation Learning
We're exploring how models build internal representations of the world, and how to engineer representations that transfer across domains. Cross-domain transfer and continual learning - acquiring new tasks without catastrophic forgetting - are core challenges we're tackling.
Mechanistic Interpretability
We're investigating attention patterns and circuit-level behaviour in transformers. The goal is to move from "it works" to "we understand why it works" - which directly informs our distillation and compression research.
The Team
Dawar - Founder. Driving the lab's vision, research agenda, infrastructure, and strategic direction.
Nilesh Sarkar - Founding Researcher. Leads knowledge distillation and LLM compression research. Previously AI & Deep Learning Researcher at Moog Controls (internship extended for high-impact delivery), building production agentic AI systems for aerospace.
Dhruv - Research Scientist. Contributing to the lab's core research on representation learning and efficient AI architectures.
What's Next
We're actively publishing our findings, building open tools for the research community, and exploring collaborations with institutions working on efficient AI. If you're interested in our work, get in touch.
Follow our progress on the Erdős AI Lab research page.