Bayesian Reranker by Quante Carlo
Home
Retrieval as a Bayesian Optimization Problem Generative Engine Optimization using Monte Carlo Tree Search
Build your own Contact us

Bayesian Reranker

        

by Quante Carlo  

This demo uses Parallel Bayesian Optimization to increase context at each iteration for RAG using the miniwiki dataset.

 
Number of results per keyword term to fetch from corpus:  

How it works

    First screen:
  1. Rewrites your prompt
  2. Comes up with search terms based on the improved prompt
  3. Performs an embedding based search (chroma db) based on the more detailed search terms
    4. Second screen:
  4. On the first iteration, the best few based on the embedding search are chosen
    There is an llm scorer in the background that rates the relevance, reliability and recenty of the retrieved documents.
  5. For each iteration after that, batch Bayesian Optimization is performed to decide what are the next best choices based on what's been learned about the retrieval's so far.