Rerankers have been a common component of retrieval pipelines for many years. They allow us to add a final "reranking" step to our retrieval pipelines — like with Retrieval Augmented Generation (RAG) — that can be used to dramatically optimize our retrieval pipelines and improve their accuracy.

In this video we'll learn about rerankers, how they compare to the more common embedding retrieval only setup, and how we can create retrieval pipelines with reranking using Cohere AI reranking model. We'll also be using the (more typical) OpenAI text-embedding-ada-002 model with the Pinecone Vector Database.

📌 Code (08:32):
https://github.com/pinecone-io/exampl...

📚 Article:
https://www.pinecone.io/learn/series/...

🌲 Subscribe for Latest Articles and Videos:
https://www.pinecone.io/newsletter-si...

👋🏼 AI Consulting:
https://aurelio.ai

👾 Discord:
  / discord  

Twitter:   / jamescalam  
LinkedIn:   / jamescalam  

00:00 RAG and Rerankers
01:25 Problems of Retrieval Only
04:32 How Embedding Models Work
06:34 How Rerankers Work
08:20 Implementing Reranking in Python
13:11 Testing Retrieval without Reranking
15:21 Retrieval with Cohere Reranking
21:54 Tips for Reranking

#artificialintelligence #nlp #ai #openai