Escalable and addressable databases infractructures

Questions
1

Hello Make Community! I have a client—an e-commerce—that already uses Make, and their data is currently structured in BigQuery. While BigQuery works, they’re looking for a more natural-language-friendly way to extract insights and interact with their data. We’re considering restructuring their database. One option is using Notion, since they already use it, and it integrates well with Make. However, I’d love to hear the community’s thoughts. Are there alternative structures or platforms you’d recommend for flexible, natural-language-driven data interaction, especially in combination with Make? Thanks in advance

1 Like
3

Hi @Doudoubcn ,

The recommended approach is a RAG (Retrieval-Augmented Generation) setup, orchestrated through Make.

Here’s how it works in simple terms: The user types a question → Make receives it → the question is converted into a vector → Pinecone finds the most relevant data → the LLM puts together a clear, plain-English answer.

How RAG handles different question types:

  • Data questions — like “What was last month’s revenue?” The relevant data is pulled from Pinecone and the LLM generates the answer from it.
  • Knowledge questions — like “What are customers complaining about?” Make searches your stored documents and the LLM puts together a clear answer.

What you need:

  • An LLM API (Claude or GPT-4)
  • A vector database (Pinecone, Supabase, or Qdrant)
  • Make connects everything

RAG works best for: product Q&A, review analysis, return policies, support history, and transactional data — all stored and retrieved from your vector database.

BigQuery → Pinecone Migration Flow

  1. Extract data from BigQuery — Pull the rows/columns you want to make searchable (e.g. revenue records, customer reviews, support tickets, product descriptions).
  2. Convert to embeddings — Pass that text through an embedding model (like OpenAI’s or Claude) — this turns each piece of text into a vector (a list of numbers that captures meaning).
  3. Push to Pinecone — Store those vectors in Pinecone with metadata (like row ID, date, category) so you can filter results later.
  4. Keep it in sync — Set up a scheduled job (via Make, Cloud Functions, or a Python script) to pull new/updated rows from BigQuery and upsert them into Pinecone regularly.

What you need:

  • BigQuery access (read-only is fine)
  • An embedding model API (OpenAI, Cohere, or Google)
  • A Pinecone account and index
  • A script or Make scenario to automate the pipeline

Best regards,
@Mohamed_Jahar