Repository files navigation

Born from a simple experiment with Claude (Anthropic's LLM), OdiaLingua emerged from a vision to make complex information accessible to Odia speakers. Inspired by the groundbreaking work of theOdiaGenAI.org community, who are pioneering Odia language model fine-tuning, I focused on developing the application layer to bridge the gap between advanced AI technology and everyday Odia users.

What started as a Streamlit prototype has evolved into a full-stack application, aiming to break down language barriers and make information accessible to all Odia speakers, regardless of their technical expertise.

OdiaLingua is not just a language learning platform – it's a comprehensive solution for making information accessible to Odia speakers. The platform enables users to:

• Ask complex questions and receive simple explanations in Odia
• Access real-time information through integrated search capabilities
• Utilize text-to-speech for better accessibility
• Engage in natural conversations with AI in Odia

• LLM Provider: Groq (hosting Llama-3.3-70b)
  • Leverages Linear Processing Unit (LPU) technology for unprecedented inference speed
  • Currently the fastest LLM inference platform available
• TTS Engine: facebook/mms-tts-ory (hosted on Hugging Face)
  • Specialized model for high-quality Odia speech synthesis
• Search Integration:
  • DuckDuckGo API for real-time information
  • SerpAPI for enhanced search capabilities

• Core: React 18 + TypeScript + Vite
• Styling: Tailwind CSS + Framer Motion
• State Management: React Context + Zustand
• Authentication: Appwrite

• API: FastAPI
• ML Operations: Transformers

• Node.js 16+
• Python 3.8+
• Appwrite instance
• Groq API key
• SerpAPI key

cd frontendnpm installnpm run dev

cd backendpython -m venv venvsource venv/bin/activate# Windows: venv\Scripts\activatepip install -r requirements.txtpython app.py

Create.env files in both frontend and backend directories:

Frontend.env:

VITE_APPWRITE_PROJECT_ID=your_project_idVITE_APPWRITE_ENDPOINT=your_endpointVITE_BACKEND_URL=http://localhost:5000

Backend.env:

GROQ_API_KEY=your_groq_api_keySERPAPI_KEY=your_serpapi_keyGROQ_MODEL=llama-3.3-70b-specdec

1. Local LLM Support

   • Integration with Ollama for local LLM deployment
   • Support for Odia fine-tuned models

2. Enhanced Accessibility

   • Speech-to-Text (STT) implementation
   • Improved UI
   • Offline mode support

3. UI Enhancements

   • Advanced chat features
   • Multi-modal support
   • Better conversation management

Currently, OdiaLingua runs locally due to hosting constraints. The TTS model requires significant GPU resources, making free-tier cloud hosting impractical. As our favorite dev joke goes: "GPU poor, but spirits rich! 😅"

We welcome contributions! Here's how you can help:

1. Code Contributions

   • Fork the repository
   • Create a feature branch
   • Submit a pull request

2. Language Support

   • Help improve Odia translations
   • Contribute to language model training

3. Documentation

   • Improve README
   • Add code comments
   • Create tutorials

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

Himanshu Mohanty

• GitHub:@HimanshuMohanty-Git24
• Portfolio:himanshumohanty.netlify.app
• LinkedIn:himanshumohanty

• Groq for providing the LLM capabilities
• Hugging Face for the VITS model
• SerpAPI for search integration
• The Odia language community for inspiration and support