KumbhLabs Research
Building the Agentic Systems
for the Kumbh Ecosystem.
We design, deploy and prove AI-agent systems that connect pilgrims, businesses, startups and services - inside the world's most demanding temporary city.
Publications
Publication
Aug 1, 2026
KumbhDoot: A Scale-Ready, LLM-Bounded Architecture for Mass-Gathering Public-Service Assistants
A similarity-first architecture that bounds generative model usage behind semantic caching, approved information and intent routing - reducing cost and latency while keeping answers grounded.
Research
Jul 29, 2026
AI Agents for Kumbh Mela
A proposal for deploying agentic AI systems at the world's largest mass gathering, built around citizen-facing assistants, shared public infrastructure and scalable population intelligence.
Previous Work
Publication
MIT Media Lab
Decentralized AI: A Perspective
A framing of decentralized AI systems and why coordination without central control matters for real-world deployment.
Preprint
arXiv
Agentic Systems at Population Scale
Research on architectures for large populations of interoperating agents operating under real-world constraints.
Project
MIT Media Lab
AI-LPM: Large Population Models
An overview of large population models - simulating and coordinating the behaviour of millions of agents.
Video Vault
KumbhLabs - Talk 01
KumbhLabs - Talk 02
KumbhLabs - Talk 03
KumbhLabs - Talk 04