omnirag 2.0.2

Multi-Language Voice RAG Framework

OmniRAG

Intelligent Retrieval-Augmented Generation with Multi-Language Support and Voice Interface

Features • Installation • Quick Start • Documentation • Examples • Contributing

---

Overview

OmniRAG is a production-ready RAG (Retrieval-Augmented Generation) framework that combines adaptive learning, intelligent tool selection, and native multi-language capabilities. Unlike traditional RAG systems, OmniRAG translates outputs after retrieval, preserving semantic quality while supporting 27+ languages with built-in voice interface.

Key Capabilities

• Adaptive Intelligence: Automatically adjusts response complexity based on user expertise level
• Multi-Language Translation: Native support for 27+ languages including Tamil, Hindi, Spanish, and more
• Voice Interface: Built-in speech-to-text and text-to-speech capabilities
• Intelligent Tool Selection: Automatically chooses between local knowledge base and web search
• Production Ready: Optimized for real-world applications with caching and error handling

---

Features

Core Features

• 🌊 Liquid RAG: Adapts responses to user expertise (beginner, intermediate, expert)
• 🤖 Agentic RAG: Intelligently selects optimal information sources
• ⛓️ Chain RAG: Decomposes and handles complex multi-part queries
• 📄 Document Processing: Native support for PDF, TXT, JSON, and more
• 🔍 Vector Search: High-performance FAISS-based similarity search
• 💾 Smart Caching: Automatic response caching for improved performance

v2.0 New Features

• 🌍 Post-Retrieval Translation

  • Preserves embedding quality by maintaining original language documents
  • Significantly more storage efficient than pre-translation approaches
  • Supports full language names ("Spanish" vs "es") for better usability

• 🎤 Native Voice Interface

  • Built-in speech recognition and synthesis
  • No external API dependencies
  • Multi-language voice support
  • Works offline

• 🔧 Production Enhancements

  • UTF-8 encoding support for non-Latin scripts
  • Improved error handling and logging
  • Comprehensive documentation and examples

---

Installation

Basic Installation

pip install omnirag

With Voice Input Support

Voice output is enabled by default. For voice input (speech recognition), install additional dependencies:

Windows

pip install pipwin
pipwin install pyaudio
pip install omnirag[voice-input]

macOS

brew install portaudio
pip install omnirag[voice-input]

Linux

sudo apt-get install portaudio19-dev python3-pyaudio
pip install omnirag[voice-input]

From Source

git clone https://github.com/Giri530/omnirag.git
cd omnirag
pip install -e .

---

Quick Start

Basic Usage

from omnirag import OmniRAG

# Initialize with your preferred model
rag = OmniRAG(model_name="google/flan-t5-small")

# Add documents
rag.add_documents([
    "Python is a high-level programming language.",
    "It emphasizes code readability and simplicity."
])

# Query the system
result = rag.query("What is Python?")
print(result['answer'])

Multi-Language Translation

from omnirag import OmniRAG

# Initialize with target language
rag = OmniRAG(
    model_name="google/flan-t5-small",
    output_language="Spanish"
)

# Add English documents
rag.add_documents([
    "Artificial Intelligence enables machines to learn from experience.",
    "Machine Learning is a subset of AI."
])

# Query in any language, receive Spanish response
result = rag.query("What is AI?")
print(result['answer'])
# Output: "La Inteligencia Artificial permite a las máquinas aprender de la experiencia."

Voice-Enabled RAG

from omnirag import OmniRAG

# Initialize with voice support
rag = OmniRAG(
    enable_voice=True,
    output_language="Tamil"
)

rag.add_documents(["Quantum computing uses quantum mechanics principles."])

# Text query with spoken response
result = rag.query("Explain quantum computing", speak_answer=True)

# Full voice interaction (requires microphone)
result = rag.voice_query()

---

Supported Languages

OmniRAG supports 27+ languages with both full names and ISO codes:

Indian Languages: Tamil, Hindi, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Bengali

European Languages: Spanish, French, German, Italian, Portuguese, Russian, Polish, Dutch, Turkish

Asian Languages: Chinese (Simplified), Japanese, Korean, Vietnamese, Thai, Indonesian, Malay

Other: Arabic, English

---

Model Support

Recommended Models

Model	Parameters	RAM	Speed	Quality	Use Case
google/flan-t5-small	80M	0.5GB	⚡⚡⚡	⭐⭐	Development, Testing
google/flan-t5-base	250M	1GB	⚡⚡⚡	⭐⭐⭐	Production (Balanced)
Qwen/Qwen2.5-0.5B-Instruct	500M	1GB	⚡⚡	⭐⭐⭐	High Quality
Qwen/Qwen2.5-1.5B-Instruct	1.5B	2GB	⚡⚡	⭐⭐⭐⭐	Best Quality
Qwen/Qwen2.5-3B-Instruct	3B	4GB	⚡	⭐⭐⭐⭐⭐	Maximum Quality

---

Advanced Configuration

rag = OmniRAG(
    model_name="google/flan-t5-base",      # LLM model
    embedding_model="all-MiniLM-L6-v2",    # Embedding model
    enable_web_search=True,                 # Enable web search
    output_language="Tamil",                # Target language
    enable_voice=True,                      # Voice interface
    use_4bit=False,                         # 4-bit quantization
    verbose=True                            # Debug logging
)

---

Examples

Document Q&A System

# Load documents from various sources
rag.load_from_file("company_handbook.pdf")
rag.load_from_folder("./policy_documents")

# Query with automatic source attribution
result = rag.query("What is the remote work policy?")
print(f"Answer: {result['answer']}")
print(f"Sources: {result['sources']}")

Multi-Language Customer Support

# Initialize for Hindi-speaking users
support_rag = OmniRAG(
    output_language="Hindi",
    enable_voice=True
)

support_rag.load_from_file("product_manual.pdf")

# Customer query with voice response
result = support_rag.query(
    "How do I reset my password?",
    speak_answer=True
)

Educational Assistant

# Initialize for students
edu_rag = OmniRAG(model_name="google/flan-t5-base")

edu_rag.load_from_file("physics_textbook.pdf")

# Automatically adapts to user level
beginner_result = rag.query("Explain photosynthesis")  # Simple explanation
expert_result = rag.query("Explain quantum entanglement")  # Technical detail

Complex Query Handling

# Automatically decomposes complex queries
result = rag.query("""
Compare the advantages and disadvantages of solar and wind energy.
Which is more cost-effective for residential use?
What are the environmental impacts of each?
""")

---

API Reference

Core Methods

__init__(**kwargs)

Initialize OmniRAG with configuration parameters.

Parameters:

• model_name (str): HuggingFace model identifier
• embedding_model (str): Sentence transformer model
• enable_web_search (bool): Enable web search capability
• output_language (str): Target language for responses
• enable_voice (bool): Enable voice interface
• use_4bit (bool): Use 4-bit quantization for memory efficiency
• verbose (bool): Enable detailed logging

query(user_query, output_language=None, speak_answer=False)

Submit a query to the RAG system.

Parameters:

• user_query (str): The question or prompt
• output_language (str, optional): Override default output language
• speak_answer (bool): Enable voice response

Returns: Dictionary containing:

• answer (str): Generated response
• sources (list): Retrieved source documents
• user_level (str): Detected expertise level
• output_language (str): Language code of response

add_documents(documents)

Add documents to the knowledge base.

Parameters:

• documents (list): List of text strings or file paths

load_from_file(file_path, chunk_size=None)

Load and process a document file.

Parameters:

• file_path (str): Path to document file
• chunk_size (int, optional): Character limit per chunk

voice_query(output_language=None)

Process voice input and provide voice output.

Parameters:

• output_language (str, optional): Override default output language

Returns: Same structure as query()

---

Architecture

┌─────────────┐
│ User Query  │
└──────┬──────┘
       │
       ▼
┌─────────────────────┐
│  Liquid Analyzer    │  ← Detect user expertise level
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Chain Decomposer   │  ← Break complex queries
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Agentic Planner    │  ← Select tools (Vector DB / Web)
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Information        │
│  Retrieval          │  ← Fetch relevant chunks
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Content            │
│  Transformation     │  ← Adapt to user level
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  LLM Generation     │  ← Generate response
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Smart Translator   │  ← Translate (if needed)
└──────┬──────────────┘
       │
       ▼
┌─────────────────────┐
│  Voice Processor    │  ← Synthesize speech (if enabled)
└──────┬──────────────┘
       │
       ▼
┌─────────────┐
│   Response  │
└─────────────┘

---

Use Cases

Enterprise Applications

• Customer Support: Multi-language support with voice interface
• Internal Knowledge Base: Quick access to company documentation
• Training Systems: Adaptive content delivery based on employee expertise

Education

• Study Assistants: Personalized explanations for students
• Language Learning: Cross-language practice and translation
• Accessibility: Voice interface for visually impaired students

Research

• Literature Review: Query across multiple papers and documents
• Data Analysis: Natural language interface to research data
• Collaborative Tools: Multi-language research team support

---

Performance Optimization

Memory Management

# Use 4-bit quantization for large models
rag = OmniRAG(
    model_name="Qwen/Qwen2.5-3B-Instruct",
    use_4bit=True  # Reduces memory usage by ~75%
)

Caching

# Automatic caching of frequent queries
result1 = rag.query("What is Python?")  # ~2s
result2 = rag.query("What is Python?")  # <10ms (from cache)

# Clear cache when needed
rag.clear_cache()

Batch Processing

questions = [
    "What is machine learning?",
    "What is deep learning?",
    "What is neural network?"
]

results = [rag.query(q) for q in questions]

---

Comparison with Other Frameworks

Feature	LangChain	LlamaIndex	OmniRAG
Built-in Translation*	❌	❌	✅
Built-in Voice I/O*	❌	❌	✅
Adaptive Responses	⚠️ Manual	⚠️ Manual	✅ Auto
Indian Languages	⚠️ External	⚠️ External	✅ Native
Beginner Friendly	⚠️ Complex	⚠️ Complex	✅ Simple
Open Source	✅	✅	✅
Free to Use	✅	✅	✅

* Built-in = Native implementation without external APIs or services

---

Troubleshooting

Common Issues

Import Errors

# Ensure proper installation
pip install --upgrade omnirag

# Verify installation
python -c "from omnirag import OmniRAG; print('Success!')"

Voice Input Issues

# Windows
pipwin install pyaudio

# macOS
brew install portaudio && pip install pyaudio

# Linux
sudo apt-get install portaudio19-dev python3-pyaudio

Out of Memory

# Use smaller model or enable quantization
rag = OmniRAG(
    model_name="google/flan-t5-small",  # Smaller model
    use_4bit=True  # Enable quantization
)

---

Contributing

We welcome contributions! Please follow these steps:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Add tests for new functionality
5. Commit changes (git commit -m 'Add amazing feature')
6. Push to branch (git push origin feature/amazing-feature)
7. Open a Pull Request

Development Setup

git clone https://github.com/Giri530/omnirag.git
cd omnirag
pip install -e ".[dev]"
pytest tests/  # Run tests

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Citation

If you use OmniRAG in your research or project, please cite:

@software{omnirag2025,
  title={OmniRAG: Intelligent Multi-Language RAG Framework},
  author={Girinath V},
  year={2025},
  version={2.0.0},
  url={https://github.com/Giri530/omnirag},
  license={MIT}
}

---

Acknowledgments

Built with these excellent open-source projects:

• Transformers by Hugging Face
• FAISS by Meta Research
• Sentence Transformers by UKP Lab
• Deep Translator
• pyttsx3

---

Support

• Documentation: Complete Guide (PDF)
• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: [e-mail]

---

Roadmap

v2.0 (Current)

• ✅ Multi-language translation
• ✅ Voice interface
• ✅ UTF-8 support

v2.1 (Planned)

• DOCX and XLSX support
• Custom translation models
• Voice language selection
• Web UI

v3.0 (Future)

• Multi-modal support (images, audio)
• Real-time translation
• Cloud deployment templates
• REST API server

---

Made with ❤️ by Girinath V

⭐ Star us on GitHub • 📦 Install from PyPI • 📖 Read the Docs