RAG with Meridian

Embed documents, index them for similarity search, and pipe the top‑k chunks into a completion — all with a single pipeline definition.

Step 1

Embeddings

Chunk documents and compute embeddings via text-embedding-3-small. Vectors land in your Meridian‑managed index.

→

Step 2

Similarity Search

At query time, embed the user prompt and run ANN over your index. Top‑k chunks return with relevance scores.

→

Step 3

Completion

Stitch retrieved chunks into the system prompt and call gpt-4o. The model answers grounded in your data.

pipeline.ts

import { Meridian } from "@meridian/sdk";

const m = new Meridian({ apiKey: process.env.MERIDIAN_KEY });

const pipeline = m.pipeline("rag-demo")
  .embed({ model: "text-embedding-3-small", source: "docs/" })
  .index({ name: "knowledge-base", metric: "cosine" })
  .retrieve({ k: 5, minScore: 0.7 })
  .complete({ model: "gpt-4o", temperature: 0.2 });

const answer = await pipeline.run("What is the refund policy?");

Start building

No credit card required · 10k embeddings free