Weighted Execution Graph

The core value Finetune provides developers is access to a Weighted Execution Graph (WEG) that stores the task lists of your agents' past executions, the weights added by our evaluations, and the human feedback provided in the feedback session.

We provide an easy way for you to deploy your latest WEG to a private cloud where your graph's data is private and secure to you and only you.

Updates after fine-tuning

After each execution, we add the task list and the weights to your graph. We will add a weight of 1 for successful executions and -1 for failed executions. At this stage, each execution has the following structure:

{
  "Task List": [
    "search",
    "summarize",
    "generate", 
    "publish",
  ],
  "Weight": 1, 
}

Feedback session

After the feedback session, we convert your feedback into notes and update each execution by adding these notes. After adding the note, we will change the weight to 0, indicating that human feedback exists (useful when performing Graph RAG). At this stage, each execution has the following structure:

{
  "Task List": [
    "generate",
    "search",
    "generate", 
    "publish",
  ],
  "Weight": 0, 
}
  "Note": "Always search before generating new content and always summarize the content after generation." 

Root access

When you run your agent in production, it will first perform Graph RAG over your execution to retrieve relevant context for your agent's core LLM to create the following task list. You will have root access to perform any CRUD operations over your graph, meaning you can add to your graph by running a fine-tuning session or add suggestions during a feedback session at any time.