Retrieval-Augmented Generation — Animate | AI Wisdom

← Back to topic
🎞 ANIMATE · Mode 3 of 5 · +100 XP— RAG Pipeline
🧬Embed
🔍Retrieve
📝Augment
🤖Generate
📤Return
🎯 Next: Test →
Step 1 / 5
Speed
Embed Query
Convert question to a 1536-dim vector💡 Analogy
Like translating your question into the same dialect as the library catalog — only then can the index find matches.
The same model that embedded your documents must embed the query — otherwise vectors live in different spaces.
Console Output
> question: "How do I rotate Azure Key Vault secrets?"
> embedding: Float32Array(1536) [0.012, -0.087, ...]
●rag.ts
Embed QueryL9–13
1import OpenAI from 'openai';
2import { Pinecone } from '@pinecone-database/pinecone';
3 
4const openai = new OpenAI();
5const pc = new Pinecone();
6const index = pc.index('docs');
7 
8export async function answer(question: string) {
9  // 1. Embed the question
10  const { data: [{ embedding }] } = await openai.embeddings.create({
11    model: 'text-embedding-3-small',
12    input: question,
13  });
14 
15  // 2. Retrieve top-k chunks
16  const { matches } = await index.query({
17    vector: embedding,
18    topK: 5,
19    includeMetadata: true,
20  });
21 
22  // 3. Build grounded prompt
23  const context = matches
24    .map((m, i) => `[#${i + 1}] ${m.metadata?.text}`)
25    .join('\n\n');
26 
27  // 4. Generate with citations
28  const res = await openai.chat.completions.create({
29    model: 'gpt-4o-mini',
30    messages: [
31      { role: 'system', content: 'Answer ONLY from context. Cite as [#N]. If unknown, say so.' },
32      { role: 'user', content: `Context:\n${context}\n\nQ: ${question}` },
33    ],
34  });
35 
36  return res.choices[0].message.content;
37}
37 lines
Timeline — 5 Steps
Embed Query
Convert question to a 1536-dim vector💡 Analogy
Like translating your question into the same dialect as the library catalog — only then can the index find matches.
The same model that embedded your documents must embed the query — otherwise vectors live in different spaces.
Console Output
> question: "How do I rotate Azure Key Vault secrets?"
> embedding: Float32Array(1536) [0.012, -0.087, ...]

Was this page helpful?

Discussion

Convert question to a 1536-dim vector

💡 Analogy

Like translating your question into the same dialect as the library catalog — only then can the index find matches.

The same model that embedded your documents must embed the query — otherwise vectors live in different spaces.

Console Output

> question: "How do I rotate Azure Key Vault secrets?"
> embedding: Float32Array(1536) [0.012, -0.087, ...]

Convert question to a 1536-dim vector

💡 Analogy

Like translating your question into the same dialect as the library catalog — only then can the index find matches.

The same model that embedded your documents must embed the query — otherwise vectors live in different spaces.

Console Output

> question: "How do I rotate Azure Key Vault secrets?"
> embedding: Float32Array(1536) [0.012, -0.087, ...]