Build a RAG pipeline with Pinecone

Prerequisites

Step-by-Step

1. 1

   Create a serverless index

   Serverless indices auto-scale and bill per query. Pick the dimension to match your embedding model (1536 for text-embedding-3-small).

   pnpm add @pinecone-database/pinecone openai

2. 2

   Initialize the client

   Use a single client instance for the lifetime of your app.

   import { Pinecone } from '@pinecone-database/pinecone';
   const pc = new Pinecone({ apiKey: process.env.PINECONE_API_KEY! });
   const index = pc.index('docs');

3. 3

   Embed and upsert

   Batch upserts in groups of 100 vectors. Always include rich metadata so you can filter at query time.

   import OpenAI from 'openai';
   const openai = new OpenAI();
   const { data } = await openai.embeddings.create({ model: 'text-embedding-3-small', input: chunks.map((c) => c.text) });
   await index.upsert(
     data.map((d, i) => ({ id: chunks[i].id, values: d.embedding, metadata: { source: chunks[i].source } }))
   );

4. 4

   Query with filters

   Metadata filters narrow results before vector similarity runs - critical for multi-tenant apps.

   const q = await openai.embeddings.create({ model: 'text-embedding-3-small', input: 'reset password' });
   const res = await index.query({ vector: q.data[0].embedding, topK: 5, filter: { source: { $eq: 'help-docs' } }, includeMetadata: true });

5. 5

   Add hybrid search

   Pinecone supports sparse-dense hybrid via SPLADE or BM25 vectors. Use it when keyword matches matter (proper nouns, codes).

6. 6

   Monitor and tune

   Watch the Pinecone dashboard for read units consumed. If costs balloon, lower topK or add a cache layer in front.

Common Pitfalls

What's Next

• ->Add a reranker to squeeze more relevance out of topK results.
• ->Set up a nightly job to re-embed docs that drifted.

More Build a RAG pipeline