LLM-Related, My Study Flow

At first, this market centered on a single high-performance LLM, but it has gradually expanded in both scope and depth toward agentic systems, agents, browser-based agent groups, and most recently even toward the OS layer. In a section that changes this quickly, it is important to have the sense and cognitive ability to distinguish what is changing from what is not. That seems especially true for the somewhat ambiguous roles of planner or PM/PO.
Human brains and perception are based on bias toward what one already knows or possesses, and meta-cognition is grounded in background knowledge. Because of that, I keep following up across related areas in order to broaden my experience with various tools and utilities. 
So as not to follow things blindly or stumble backward in confusion, I grouped the study history I had gone through into categories as a kind of early? or middle? checkpoint, and organized the keywords I had encountered and practiced in sequence. 
 
 
NLP
ML -> ... -> DL -> ... -> RNN -> encoder, decoder -> LSTM -> Attention -> transformer -> LLM 
 
Cloud service
openai -> multion -> claude -> nomic -> Groq(LPU chip) -> grok-1 -> Perplexity -> arc-search(_ARC)
 
Local based, Unified Interface
ollama(webUI) -> lm studio, anything LLM -> codeGPT -> Msty
 
RAG
langchain -> agentic rag -> LLocalSearch -> SearXNG -> open parse -> phidata 
 
Agent
MOE -> devika -> crewai -> AutoGen(, ..Studio) -> langgraph(, langsmith) -> aiexe ,TaskingAI -> llm os(_phidata)
 
 
 
As mentioned at the beginning, the keywords in the work I am most recently practicing are Agentic Local LLM, WebOS, and LLM OS. 
To deepen my understanding of Agentic LLM, I am going back to the langgraph part from the beginning and studying it again. Alongside that, I am also following and trying out updates related to the experimental WebOS, a browser-based LLM, and LLM OS, a new operating-system kernel process.