Files
Obsidian/MyselfDev/langchain/AI Learning Series — Part 4Embeddings, ChromaDB, Chunking & Retrieval (With a PDF → Chroma →….md
T
coco 8b94d76799 a
2026-07-21 13:57:15 +08:00

4.6 KiB
Raw Blame History

title, source, author, published, created, description, tags
title source author published created description tags
AI Learning Series — Part 4:Embeddings, ChromaDB, Chunking & Retrieval (With a PDF → Chroma →… https://medium.com/@vasanthancomrads/ai-learning-series-part-4-embeddings-chromadb-chunking-retrieval-with-a-pdf-chroma-abc5a98ecb5f
Vasanthan K
2025-11-24 2026-04-06 Master RAG essentials with embeddings, ChromaDB, and retrieval techniques. Follow a beginner-friendly walkthrough to build a PDF-to-Chroma RAG app.
clippings

Retrieval-Augmented Generation (RAG) is one of the most practical AI patterns used today. Whether you’re building chatbots, internal knowledge assistants, or customer-facing AI apps, you need to understand embeddings, chunking, vector databases and retrieval.

In this blog, we’ll cover the core concepts step-by-step:

  • Embeddings
  • ChromaDB setup
  • Chunking strategy (50–200 tokens)
  • Retrieval pipeline
  • Exercise: Build a complete PDF → Chroma → Answer RAG app in Python

🔍 1. What Are Embeddings?

Embeddings are numerical representations of text.
They convert sentences into high-dimensional vectors (arrays of floats) that capture meaning.

Example:

"Apple the fruit" → [0.12, -0.93, 0.04, … ]
"Apple the company" → [1.02, 0.55, -0.31, … ]

Similar meanings → similar vectors.
Different meanings → distant vectors.

This is the foundation of semantic search and RAG.

⚙️ Generating Embeddings (Python Example)

from openai import OpenAI
client = OpenAI()

text = "AI helps developers build faster."
emb = client.embeddings.create(
    model="text-embedding-3-small",
    input=text
)
print(len(emb.data[0].embedding))  # Dimensionality

🗂️ 2. ChromaDB Setup

ChromaDB is a lightweight vector database ideal for RAG apps.

Install

pip install chromadb pypdf openai

Initialize a collection

import chromadb
client = chromadb.Client()

collection = client.create_collection(
    name="docs",
    metadata={"hnsw:space": "cosine"}  # cosine similarity
)

✂️ 3. Chunking Strategy (50–200 Tokens)

Large documents need to be chunked before embedding.

Why chunk?

  • Embedding long text → loses precision
  • Smaller chunks → better retrieval
  • Great trade-off size: 50–200 tokens

Simple chunking example

def chunk_text(text, chunk_size=150):
    words = text.split()
    for i in range(0, len(words), chunk_size):
        yield " ".join(words[i:i + chunk_size])

🔄 4. Retrieval Pipeline

Here’s the basic RAG flow:

🧪 Exercise: Build a “PDF → Chroma → Answer” RAG App

Let’s build a fully working pipeline.

Step 1: Load and Chunk the PDF

from pypdf import PdfReader

def load_pdf(path):
    reader = PdfReader(path)
    text = ""
    for page in reader.pages:
        text += page.extract_text() + "\n"
    return text
pdf_text = load_pdf("sample.pdf")
chunks = list(chunk_text(pdf_text, chunk_size=120))
print("Total chunks:", len(chunks))

Step 2: Embed and Insert into Chroma

from openai import OpenAI
client = OpenAI()

embeddings = client.embeddings.create(
    model="text-embedding-3-small",
    input=chunks
)
vectors = [e.embedding for e in embeddings.data]
collection.add(
    documents=chunks,
    embeddings=vectors,
    ids=[f"id_{i}" for i in range(len(chunks))]
)

Step 3: Retrieval Function

def retrieve(query, top_k=3):
    q_emb = client.embeddings.create(
        model="text-embedding-3-small",
        input=query
    ).data[0].embedding

results = collection.query(
        query_embeddings=[q_emb],
        n_results=top_k
    )
    return results["documents"][0]

Step 4: Generate Final Answer Using Retrieved Context

def answer(query):
    context_chunks = retrieve(query)
    context = "\n".join(context_chunks)

response = client.chat.completions.create(
        model="gpt-4.1-mini",
        messages=[
            {"role": "system", "content": "You are a helpful assistant for answering based on context."},
            {"role": "user", "content": f"Context:\n{context}\n\nQuestion:\n{query}"}
        ]
    )
    return response.choices[0].message["content"]
print(answer("What is the main idea of the document?"))

🎉 Final Output: A Working RAG App

You now have a complete:

PDF → Chunk → Embed → ChromaDB → Retrieve → Answer

pipeline.

This architecture is the foundation for:

  • chat-with-PDF apps
  • enterprise knowledge bots
  • Q&A search tools
  • internal documentation assistants