Forem: Tomo

Simplifying RAG with Elegant Abstractions

Tomo — Mon, 24 Mar 2025 14:59:24 +0000

Building Retrieval-Augmented Generation (RAG) applications typically involves setting up vector databases, managing embedding models, implementing chunking strategies, and writing complex queries. CapybaraDB offers an elegant abstraction layer that simplifies this entire workflow.

The Technical Challenge of RAG

RAG systems must:

Process and chunk text documents
Generate embeddings for each chunk
Store vectors efficiently
Handle semantic searches
Manage the retrieval process

Traditional implementations require separate components for each step, complicating the developer experience.

CapybaraDB's Approach

CapybaraDB consolidates these complexities through its EmbJSON concept - an extension of JSON that handles embeddings automatically.

EmbText: Automatic Vectorization

The EmbText type handles text chunking and embedding behind the scenes:

from capybaradb import CapybaraDB, EmbText

# Initialize client
client = CapybaraDB()
db = client.db("knowledge_base")
collection = db.collection("technical_articles")

# Create complex document with embedded text
article = {
    "metadata": {
        "title": "Advanced Vector Database Architecture",
        "author": {
            "name": "Jane Smith",
            "email": "jane@example.com"
        },
        "tags": ["databases", "vector search", "RAG"],
        "published": True,
        "views": 1250
    },
    "content": EmbText("""
    Vector databases have emerged as critical infrastructure for AI applications.
    They enable efficient similarity search across high-dimensional vector spaces.

    Modern vector databases must handle:
    - High throughput ingestion
    - Low-latency search
    - Scalable storage solutions
    - Advanced filtering capabilities

    This article explores architectural approaches to these challenges.
    """),
    "sections": [
        {
            "title": "Data Structures",
            "content": EmbText("HNSW and IVF are common indexing structures...")
        },
        {
            "title": "Scaling Strategies",
            "content": EmbText("Horizontal sharding distributes vector indices...")
        }
    ]
}

# Insert document - chunking and embedding happen automatically
response = collection.insert([article])

What's happening under the hood:

The document is processed recursively
Any EmbText field is automatically chunked
Chunks are embedded using the specified model
Both original text and vectors are stored

Fine-Tuning the Process

Need control over chunking? CapybaraDB provides options:

from capybaradb import EmbText, EmbModels

section = {
    "title": "Technical Deep Dive",
    "content": EmbText(
        text="Long technical content with code examples...",
        emb_model=EmbModels.TEXT_EMBEDDING_3_LARGE,
        max_chunk_size=500,
        chunk_overlap=50,
        separators=["\n\n", "\n", ". "]
    )
}

Performing Semantic Search

Retrieval is equally straightforward:

# Basic semantic search
results = collection.query({
    "query": "How do vector databases handle scaling?",
    "embedding_model": "text-embedding-3-small",
    "top_k": 3
})

# Access the semantically relevant chunks
for match in results["matches"]:
    print(f"Score: {match['score']}")
    print(f"Matching text: {match['chunk']}")
    print(f"Document path: {match['document_path']}")

Combining Traditional and Semantic Queries

CapybaraDB allows mixing traditional filtering with semantic search:

# Find articles about databases with >1000 views that mention scaling
results = collection.query(
    {
        "metadata.tags": "databases",
        "metadata.views": {"$gt": 1000},
        "query": "scaling vector databases",
        "top_k": 5
    }
)

Technical Benefits

Separation of concerns: Application code deals with documents; CapybaraDB handles vectorization
Automatic chunking: Text is split according to your specifications
Context retention: Each chunk knows its source document and location
Query flexibility: Combine metadata filtering with semantic search
Model agnostic: Switch embedding models without changing your code

Implementation Architecture

CapybaraDB's implementation:

Uses a document-first schema design
Maintains referential integrity between chunks and source documents
Indexes metadata fields for efficient filtering
Optimizes vector storage for rapid similarity search
Handles the complexity of recursive document processing

This architecture allows developers to focus on building RAG applications rather than managing vector infrastructure.

Conclusion

CapybaraDB's elegant abstraction demonstrates how technical complexity can be hidden behind thoughtful interfaces. By treating embeddings as first-class citizens in a document database, it bridges the gap between traditional data models and vector-based retrieval systems.

For developers building RAG applications, this means less time spent on vector database management and more time creating value through AI-powered features.

Build RAG 10X Faster

Tomo — Mon, 23 Dec 2024 04:06:28 +0000

(4 min read)

1. Install CapybaraDB

pip

pip install capybaradb

npm

npm install capybaradb

2. Set your secrets

CAPYBARA_API_KEY="your_api_key"
CAPYBARA_PROJECT_ID="your_project_id"

Those secrets can be found at CapybaraDB

3. Initialize the client

Javascript

import { CapybaraDB} from "capybaradb";
import dotenv from "dotenv";

dotenv.config();

const client = new CapybaraDB();
const db = client.db("test");
const collection = db.collection("test");

Python

from capybaradb import CapybaraDB
from dotenv import load_dotenv

load_dotenv()

capybara = CapybaraDB()
db = capybara.db("test")
collection = db.collection("test")

4. Prepare example data to save

Example data set

You can either use your own data or use this example data set below.

Here we use a simple data set of famous cities, Paris, Tokyo, NYC, Rio de Janeiro.

You can add as many metadata as you want.

import {EmbText} from "capybaradb";
const docs = [
  {
    city: "Paris",
    bio: new EmbText(
      "Known as the 'City of Light,' Paris is celebrated for its romantic ambiance, iconic landmarks like the Eiffel Tower and Notre-Dame Cathedral, and world-renowned art museums such as the Louvre. The city is a hub of fashion, cuisine, and culture, with charming cafes and picturesque streets that captivate visitors."
    ),
  },Ï
  {
    city: "Tokyo",
    description: new EmbText(
      "A dazzling fusion of tradition and modernity, Tokyo offers skyscrapers, neon-lit streets, and centuries-old temples. The city is famous for its cutting-edge technology, vibrant pop culture, and culinary delights, including sushi and ramen. Tokyo is also a gateway to traditional Japanese customs, such as tea ceremonies and cherry blossom festivals."
    ),
  },
  {
    city: "New York City",
    description: new EmbText(
      "Known as 'The Big Apple,' New York City is a global center for finance, entertainment, and art. Iconic landmarks include Times Square, the Statue of Liberty, and Central Park. The city's diverse neighborhoods, from Manhattan to Brooklyn, are home to a melting pot of cultures and cuisines."
    ),
  },
  {
    city: "Rio de Janeiro",
    description: new EmbText(
      "Set against a backdrop of lush mountains and stunning beaches, Rio is famous for its vibrant Carnaval celebrations, samba music, and the Christ the Redeemer statue. The city offers breathtaking views from Sugarloaf Mountain and is a paradise for beach lovers at Copacabana and Ipanema."
    ),
  },
];

Python

from capybaradb import EmbText
docs = [
  {
    city: "Paris",
    bio: EmbText(
      "Known as the 'City of Light,' Paris is celebrated for its romantic ambiance, iconic landmarks like the Eiffel Tower and Notre-Dame Cathedral, and world-renowned art museums such as the Louvre. The city is a hub of fashion, cuisine, and culture, with charming cafes and picturesque streets that captivate visitors."
    ),
  },Ï
  {
    city: "Tokyo",
    description: EmbText(
      "A dazzling fusion of tradition and modernity, Tokyo offers skyscrapers, neon-lit streets, and centuries-old temples. The city is famous for its cutting-edge technology, vibrant pop culture, and culinary delights, including sushi and ramen. Tokyo is also a gateway to traditional Japanese customs, such as tea ceremonies and cherry blossom festivals."
    ),
  },
  {
    city: "New York City",
    description: EmbText(
      "Known as 'The Big Apple,' New York City is a global center for finance, entertainment, and art. Iconic landmarks include Times Square, the Statue of Liberty, and Central Park. The city's diverse neighborhoods, from Manhattan to Brooklyn, are home to a melting pot of cultures and cuisines."
    ),
  },
  {
    city: "Rio de Janeiro",
    description: EmbText(
      "Set against a backdrop of lush mountains and stunning beaches, Rio is famous for its vibrant Carnaval celebrations, samba music, and the Christ the Redeemer statue. The city offers breathtaking views from Sugarloaf Mountain and is a paradise for beach lovers at Copacabana and Ipanema."
    ),
  },
];

5. Save your data (No embedding needed!)

Javascript

const response = await collection.insert(docs);

Python

response = collection.insert(docs)

6. Search doc semantically (No embedding needed!)

Javascript

const query = "Global city iconic landmarks, cultural diversity, finance, entertainment, art"

const queryResult = await collection.query(query)

Python

query = "Global city iconic landmarks, cultural diversity, finance, entertainment, art"

query_result = rescollection.query(query)

Expected response

{
  matches: [
    {
      chunk: "Known as 'The Big Apple,' New York City is a global center for finance, entertainment, and art. Iconic landmarks include Times Square, the Statue of Liberty, and Central Park. The city's diverse",
      path: 'bio',
      chunk_n: 0,
      score: 0.637317419,
      document: {ObjectId("6764051ace073a82cc8ab6b2")}
    },
    {
      chunk: 'A dazzling fusion of tradition and modernity, Tokyo offers skyscrapers, neon-lit streets, and centuries-old temples. The city is famous for its cutting-edge technology, vibrant pop culture, and',
      path: 'bio',
      chunk_n: 0,
      score: 0.562242568,
      document: {ObjectId("6864051ace073a82cc8ab6de")}
    },
    {
      chunk: "The city's diverse neighborhoods, from Manhattan to Brooklyn, are home to a melting pot of cultures and cuisines.",
      path: 'bio',
      chunk_n: 1,
      score: 0.508277535,
      document: {ObjectId("7764051ace073a82cc8ab6js")}
    }
  ]
}

More…

Now you know how to retrieve relevant chunks of your document using a simple query.

To implement the RAG feature, you can simply include these chunks in your prompt when calling LLM APIs.

Also, check out CapybaraDB Docs for more details about this library.

Forem: Tomo

Simplifying RAG with Elegant Abstractions

The Technical Challenge of RAG

CapybaraDB's Approach

EmbText: Automatic Vectorization

Fine-Tuning the Process

Performing Semantic Search

Combining Traditional and Semantic Queries

Technical Benefits

Implementation Architecture

Conclusion

Build RAG 10X Faster

1. Install CapybaraDB

2. Set your secrets

3. Initialize the client

4. Prepare example data to save

Example data set

5. Save your data (No embedding needed!)

6. Search doc semantically (No embedding needed!)

Expected response

More…

Happy building!!