This project is an LLM-powered Retrieval-Augmented Generation (RAG) chatbot designed specifically for Hinglish (Hindi-English code-switched) conversations. Inspired by TurboML’s mission to democratize AI, this chatbot leverages efficient retrieval and lightweight inference to enable real-time, multilingual AI applications.
🔹 FAISS-based Retrieval – Quickly fetches relevant Hinglish text to enhance context.
🔹 LLM-powered Response Generation – Uses Mistral-7B/LLaMA-2-7B for contextual replies.
🔹 Optimized for Low-Resource Languages – Custom embeddings & reinforcement learning improve retrieval.
🔹 Efficient Inference Pipeline – Runs on GPU for fast processing in real-world applications.
🔹 Gradio-powered UI – Interactive chatbot interface for seamless deployment.
1️⃣ Retrieval: FAISS searches the most relevant Hinglish text snippets.
2️⃣ Contextual Prompting: Retrieved text is formatted into a structured prompt.
3️⃣ Response Generation: A fine-tuned LLM (Mistral/LLaMA-2) generates an appropriate response.
4️⃣ User Interaction: The chatbot provides real-time answers in Hinglish.
# Clone the repository
git clone https://github.com/LLM_Hinglish/hinglish-rag-chatbot.git
cd hinglish-rag-chatbot
# Install dependencies (Ensure GPU support)
pip install -r requirements.txt
# Run the chatbot
python app.py To launch the chatbot UI:
python app.py This will generate a public shareable link for easy access.
✅ India-first LLM Adaptation: Hinglish is widely spoken but underrepresented in AI models.
✅ Scalable & Efficient: Works well even in low-resource environments.
✅ TurboML Inspiration: Aligns with the mission of building open-source foundation models for India.
🔥 Want to contribute? Fork the repo & raise a PR!
📩 Interested in AI research? Connect with me on LinkedIn!
Let’s build India’s AI future together! 🚀 🇮🇳