18, Founder of Mantle (YC F25), ZFellow, Neovim enthusiast

Previously:

• Youngest Google grantee at 17. I used the GCP grant to build Gakudan, a multi-agent protocol that let agents:
  • Think of a question from different perspectives
  • Discuss, see other opinions from self building RAG
  • Conclude agentically
• Hedge fund data processing: built options order-book pipelines over TBs of data. Sped up the old system ~8x.
• My Neovim config grew to 280+ stars. It was the #2 biggest Neovim config until a month after release.

Tech from ideas I tried but that didn't pan out

1. Our own voice AI server, with latency similar to a local LiveKit setup
2. Weighted cosine similarity search: cluster-based weighting of vector dimensions for query vectors
3. A collaborative multi-agent protocol for agent-native knowledge systems
   • Agents discuss millions of tokens of context to find the best shape and structure for knowledge
   • Once the structure is decided, they write, iterate, and collaborate on the context
   • This worked well but was very token-expensive. It was inspired by Gakudan.

I am always happy to talk about anything and everything interesting. I am especially interested in new approaches to RL and knowledge representation.

email: samrath at mantle dot work
twitter: @samrathchadha
linkedin: in/samrathchadha