***********
        ****************
      *******************
      ********************
       ********************
          \   //  ********
           \\//  *******
             \\\////
              |||//                       ,
              |||||                    __/
  ,,,,,,,,,,,//||||\,,,,,,,,,,,,,,,,,,o==o
  ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

Recent Writing

LLM Ops

1 min read

Traditional ML: output score only Agent Learning: output (score, guidance)

  1. LLM
  2. Domain Expert

Closing the automation Loop

  • How agents are evolving
  • The persistent agent

Before LLM we always need the data. ChatGPT is innovation and revolutionary

But the problem is most of them are Silos

Siloed Task: One agent for HR, finance

Chat bot: chat bot will stop talking if we stop talking

The evolution of AI agents

  • Coding Agents
  • Developer Agents
  • App building agents (devin & Lovable **)
  • Agent Contractor ( copilot and claude code review )
  • Agent Coworker
  • self identity
  • learns from feedback

Context length crisis

  • LLM suffer from a “sensory overload”
  • We can help use vector databases to keeps the context in check ( build RAG )

Persistent Agent

  • Event driven systems
  • Modular system
  • Hirrwrchical agents

Graph View