n8n AI Agent + Cohere 3.5 Reranker #n8n #aiagents #rag #supabase

Key Concepts:

• Re-ranking: A technique to improve AI agent accuracy by filtering vector store results.
• Coher's 3.5 Re-ranker Model: A specific re-ranking model natively supported in n8n.
• Vector Store: A database used to store and retrieve data chunks for AI agents.
• Hybrid Search: Combining different search methods (e.g., keyword and semantic).
• Metadata Filtering: Filtering data based on associated metadata.
• RAG (Retrieval-Augmented Generation): An AI framework that combines information retrieval with text generation.

Setting Up Re-ranking in n8n

1. Prerequisites:

   • Ensure n8n is updated to the latest stable version (1.98 or later).
   • Have an AI agent configured to use a vector store as a tool for accessing a knowledge base.

2. Enabling Re-ranking:

   • Open the AI agent in n8n.
   • Locate and enable the "Re-rank results" toggle.
   • A "Re-ranker" leg will appear, allowing the addition of a sub-node.

3. Configuring Coher Re-ranker:

   • Click the "+" button in the "Re-ranker" leg.
   • Select "Coher" as the re-ranker.
   • Click "Create new credential" to add your Coher API key.

4. Obtaining Coher API Key:

   • Go to coher.com and create an account.
   • In the Coher dashboard, navigate to "API Keys."
   • Generate a new trial key and copy it.

5. Adding Coher API Key to n8n:

   • Paste the copied Coher API key into the credential field in n8n.
   • Click "Save."

6. Selecting Re-ranker Model:

   • Choose the desired Coher re-ranker model. The video uses version 3.5.

Example: Question and Response

• Question: "What rules govern the power unit of an f1 car?"
• The question is sent to the AI agent.
• The agent queries the vector store, retrieving relevant chunks.
• The retrieved chunks are sent to the Coher re-ranker.
• Coher's ranker identifies the most relevant results from the knowledge base.
• The agent generates a response grounded in these re-ranked results.

Benefits of Re-ranking

• Improved Accuracy: Re-ranking significantly enhances the accuracy of AI agent outputs compared to standard semantic search. Benchmarks demonstrate this improvement.
• Fine-tuned Responses: Combining re-ranking with techniques like dynamic metadata filtering and hybrid search allows for highly accurate and tailored responses.

Real-World Application: RAG Agent Workflow

• The video references a RAG agent workflow in the n8n community.
• This workflow combines re-ranking with dynamic metadata filtering and hybrid search.
• The result is a system that provides very fine-tuned and accurate responses.

Notable Quotes:

• "Ranking is a great technique to improve the accuracy of an ai agent by only sending the most relevant chunks returned from a vector store to the agent when formulating a response."
• "There are benchmarks that show that this technique can dramatically improve the outputs of an ai agent compared to standard semantic search."

Conclusion:

Re-ranking, particularly using Coher's models within n8n, is a powerful technique to enhance the accuracy and relevance of AI agent responses. By filtering the results from a vector store, re-ranking ensures that the agent focuses on the most pertinent information when generating its output. Combining re-ranking with other techniques like hybrid search and metadata filtering can further refine the agent's performance, leading to more accurate and contextually appropriate answers.