Decision tree for building a custom LLM solution

graph.viz

digraph G {
  graph [fontname = "Handlee"];
  node [fontname = "Handlee"];
  edge [fontname = "Handlee"];

  bgcolor=transparent;

  subgraph cluster_0 {
    color=orange;
    label="Usage"
    node [shape=box];
    needs[label="What are the specific needs"];
    know[label="Knowledge retrieval"];
    agents[label="Agents"];
    general[label="General"];
    prompt[label="Prompt engineering"];
    fewshot[label="Few-shot learning"];
    train[label="Fine-tuning"];
    RAG;
    SFT;
    RLHF;
    
    needs -> know;
    needs -> agents;
    needs -> general;
    
    know -> prompt;
    agents -> prompt;
    general -> prompt;
    
    prompt -> RAG;
    prompt -> fewshot;
    prompt -> train;
    
    train -> SFT;
    train -> RLHF;
  }

  subgraph cluster_1 {
    label="Deployment";
    color=blue
    node [shape=box];
    security[label="What are the security constraints?"];
    cloud[label="Cloud provider"];
    local[label="Local deployment"];
    hardware[label="What's the hardware setup?"];

    security -> cloud;
    security -> local;
    
    cloud -> OpenAI;
    cloud -> Azure;
    cloud -> Other;
    
    local -> hardware;
    hardware -> CPU;
    hardware -> GPU;
    hardware -> Custom;
  }
  
  start[label="Building a Custom LLM"];
  start -> needs;
  start -> security;

  start [shape=box];
}