Umar Khatana

Peer-to-peer ranking protocol for AI models

decentralised AI compute at Telegraph

Senior Engineer, Telegraph · 2026 · AI, Blockchain

Try Alexandria → Telegraph

The Telegraph home page: a peer-to-peer ranking protocol for machine intelligence

The problem

A decentralised network ranking AI models needs validator and miner nodes to agree on state without trusting each other, and to prove that state on-chain.

Outcome

Live, powering Alexandria

Stack

  • Go
  • Solidity
  • gRPC
  • EVM
  • P2P networking

What I built

Telegraph is a peer-to-peer ranking protocol for machine intelligence: anything behind an API can plug in and compete per task, and demand is routed to the best intelligence for that specific job. Alexandria is the app people use to ask.

I write the Go node software for subnet synchronisation and peer-to-peer networking across validators and miners, and the Solidity contracts for cross-chain state verification, so the network agrees on how models rank without trusting any single party and proves that ranking on-chain. I profiled and benchmarked the nodes under load to remove throughput bottlenecks.

Telegraph’s demand-and-supply diagram, with Alexandria as the demand side
Requests flow from humans, apps and agents; miners answer; Telegraph keeps a live ranking and routes the next request to whoever is winning.
The Telegraph architecture: task markets, validators and the economic loop
Per-task markets of miners and evaluators, finalised by validators, inside an economic loop that pays for better intelligence.