Lecture Notes: Creating AI Agents with NA10 and Vector Databases

Introduction

• Speaker: Alexander, AI automation specialist
• Objective: Export a database into a vector database to create AI agents
• AI agents will use this database to answer questions

Setting Up a Project in NA10

• Start a new project
• Create a trigger
• Fetch data from a database, e.g., property listing

Data Handling

• Example: 100 records in a table
• Create an API request to grab records
  • Use Airtable API
  • Create HTTP request and import curl
  • Obtain URL with base ID and table name
  • Extract base ID and table ID

Token Creation

• Create a token in the builder hub
• Provide access to the database
• Scope: add all records and base scheme

Extracting Data

• Test data extraction
• Extract records from array

Creating a Vector Store

• Use Pinecone API
• Export data to a file (optional)
• Create Pinecone vector store from data
  • Requires a Pinecone API account and key
  • Create index with text embedding
  • Name index (e.g., "property")

Data Embedding

• Select text embedding model
• Create JSON file
• Form a sentence with all data fields for storage
  • Example sentence: Property name, address, etc.

Testing and Finalizing

• Test workflow
• Update table with sentences (chunks)
• Each chunk corresponds to a record

Creating an AI Agent

• Add a trigger for a chat AI agent
• Use OpenAI for memory
• Define context for dialogue
• Use Vector Store Question Answer Tool
  • Select Pinecone vector store
• Select OpenAI model for chat

Example Interaction

• Start chat
• Example question: "What is the most expensive property in the list?"
• AI checks the vector store and provides the answer

Conclusion

• Demonstrated approach to building a vector database for AI agents