Skip to main content
This guide walks you through creating a complete semantic search pipeline using VectorDB with Chroma as the vector database.

​
What you’ll build

By the end of this quickstart, you’ll have:
  • A working semantic search pipeline using Chroma
  • Documents indexed with dense embeddings
  • The ability to search using natural language queries
  • Optional RAG answer generation
This example uses Chroma for local development. You can swap to any other supported vector database (Pinecone, Weaviate, Qdrant, Milvus) by changing the configuration.

​
Before you begin

Make sure you’ve completed the installation steps and have:
  • Python 3.11+ installed
  • VectorDB dependencies installed via uv sync
  • Virtual environment activated

​
Step 1: Create a configuration file

Create a configuration file config.yaml that defines your search pipeline:
config.yaml
dataloader:
  type: "arc"
  split: "test"
  limit: 100
  use_text_splitter: false

embeddings:
  model: "sentence-transformers/all-MiniLM-L6-v2"
  device: "cpu"
  batch_size: 32

chroma:
  collection_name: "quickstart-semantic-search"
  path: "./chroma_data"
  recreate: true

search:
  top_k: 5

rag:
  enabled: false

logging:
  name: "vectordb_quickstart"
  level: "INFO"
This configuration uses the ARC dataset (science reasoning questions) and loads 100 test documents. The embedding model runs on CPU for compatibility.

​
Step 2: Index your documents

Create a Python script index.py to index documents into Chroma:
index.py
from vectordb.langchain.semantic_search import ChromaSemanticIndexingPipeline

# Initialize the indexing pipeline
pipeline = ChromaSemanticIndexingPipeline("config.yaml")

# Run the indexing process
result = pipeline.run()

print(f"Successfully indexed {result['documents_indexed']} documents")
print(f"Collection: {result['collection_name']}")
Run the indexing script: 
python index.py
You should see output like: 
Successfully indexed 100 documents
Collection: quickstart-semantic-search
Indexing creates dense vector embeddings for each document using the specified model. These embeddings capture semantic meaning for similarity search.

​
Step 3: Search your documents

Create a search script search.py to query your indexed documents:
search.py
from vectordb.langchain.semantic_search import ChromaSemanticSearchPipeline

# Initialize the search pipeline
pipeline = ChromaSemanticSearchPipeline("config.yaml")

# Run a semantic search query
results = pipeline.search(
    query="What is photosynthesis?",
    top_k=5
)

print(f"Query: {results['query']}")
print(f"\nTop {len(results['documents'])} results:\n")

for i, doc in enumerate(results['documents'], 1):
    score = doc.metadata.get('score', 'N/A')
    content = doc.page_content[:200]  # First 200 characters
    print(f"{i}. [Score: {score}]")
    print(f"   {content}...\n")
Run the search script: 
python search.py
You’ll see the top 5 most semantically similar documents to your query.

​
Step 4: Add RAG answer generation (optional)

To generate answers from retrieved documents, enable RAG in your configuration:
config.yaml
rag:
  enabled: true
  model: "llama-3.3-70b-versatile"
  api_key: "${GROQ_API_KEY}"
  temperature: 0.7
  max_tokens: 2048
Make sure you have set your GROQ_API_KEY environment variable before enabling RAG.
Update your search script to display the generated answer:
search.py
from vectordb.langchain.semantic_search import ChromaSemanticSearchPipeline

pipeline = ChromaSemanticSearchPipeline("config.yaml")

results = pipeline.search(
    query="What is photosynthesis?",
    top_k=5
)

print(f"Query: {results['query']}\n")

# Display RAG-generated answer
if 'answer' in results:
    print(f"Answer: {results['answer']}\n")

print(f"Retrieved {len(results['documents'])} documents")
Now when you run the search, you’ll get an AI-generated answer based on the retrieved documents.

​
Understanding the pipeline

Here’s what happens under the hood:
1

Query embedding

Your query text is converted to a dense vector using the same embedding model used during indexing
2

Similarity search

The vector database finds documents with embeddings closest to your query embedding using cosine similarity
3

Result ranking

Results are ranked by similarity score, with the most relevant documents returned first
4

Optional RAG generation

If enabled, the LLM generates an answer using the retrieved documents as context

​
Try different vector databases

VectorDB supports multiple vector databases with the same interface. Here’s how to switch from Chroma to Pinecone: 
from vectordb.langchain.semantic_search import ChromaSemanticSearchPipeline

pipeline = ChromaSemanticSearchPipeline("config.yaml")
results = pipeline.search("What is machine learning?", top_k=5)
Just update your config file with the appropriate database settings and API keys.

​
Using Haystack instead of LangChain

VectorDB provides identical functionality for both frameworks: 
from vectordb.langchain.semantic_search import ChromaSemanticSearchPipeline

pipeline = ChromaSemanticSearchPipeline("config.yaml")
results = pipeline.search("What is photosynthesis?", top_k=5)

for doc in results["documents"]:
    print(doc.page_content)

​
Next steps

Now that you have a working semantic search pipeline, explore advanced features:

Hybrid search

Combine dense and sparse retrieval for better results

Reranking

Use cross-encoders for higher precision

Query enhancement

Improve recall with multi-query and HyDE

Metadata filtering

Filter results by document attributes

Build docs developers (and LLMs) love

Get started for freeTalk to us