Virtual-Twin

🤖 Virtual Tuhin Kumar Dutta — AI Digital Twin

Welcome to Virtual Tuhin Kumar Dutta, an intelligent Retrieval-Augmented Generation (RAG) chatbot that embodies the digital essence of Tuhin Kumar Dutta. This AI-powered assistant is designed to answer your questions about Tuhin’s expertise, background, and digital footprint — bringing his knowledge and persona to life interactively.

📌 Overview

In an era where personalized AI is transforming how we connect and consume information, this project offers a sophisticated yet user-friendly chatbot that serves as a virtual extension of Tuhin Kumar Dutta. Combining state-of-the-art NLP models, vector search, and large language models, the chatbot retrieves relevant knowledge dynamically to deliver precise and contextual responses.

🚀 Key Features

• RAG-Based Architecture: Seamlessly blends document retrieval with generative AI to provide accurate, context-aware answers.
• Sentence Embeddings: Utilizes the all-mpnet-base-v2 model to convert textual knowledge into dense vector representations for efficient semantic search.
• Vector Database: Powered by FAISS to perform lightning-fast similarity search on large knowledge bases.
• Large Language Model: Integrates the powerful llama-3.3-70b-versatile LLM, enabling rich, coherent, and human-like conversations.
• Streamlit UI: A sleek, interactive

⚙️ Technology Stack

Component	Details
Embedding Model	all-mpnet-base-v2 Sentence Transformer
Vector Database	FAISS (Facebook AI Similarity Search)
Language Model	LLaMA 3.3 70B Versatile
Frontend UI	Streamlit App
Deployment Platform	Hugging Face Spaces

📤 Key Deployment Highlights

This project is deployed on Hugging Face Spaces using a custom Docker container, enabling complete control over dependencies, performance optimization, and caching.

• Dockerized Streamlit App: The application is containerized to ensure consistent behavior across environments and to allow custom configurations not possible in standard Spaces runtime.
• Base Image and Setup: Uses the lightweight python:3.9-slim image, along with essential build tools and Git to support package compilation and version control.
• Efficient Caching: Environment variables such as HF_HOME, TRANSFORMERS_CACHE, and SENTENCE_TRANSFORMERS_HOME are explicitly defined and pointed to a shared cache directory with full read/write permissions to optimize model loading and reduce cold starts.
• Groq LLM API Integration: The chatbot utilizes the LLaMA 3.3 70B Versatile model served through Groq’s ultra-low-latency inference engine. The Groq API key is securely integrated during runtime, making responses lightning-fast while keeping the deployment architecture lean.
• Health Monitoring: A HEALTHCHECK directive pings the Streamlit server to confirm successful boot and uptime — useful for debugging issues within the Hugging Face Space.
• Port and Entry Point: The container exposes port 8501, and the entry point starts the Streamlit app with proper network bindings, making it fully accessible once deployed.

🛠️ How It Works

1. User Query: The user submits a question via the chatbot interface.
2. Embedding: The query is transformed into an embedding vector using all-mpnet-base-v2.
3. Search: FAISS searches the vector database for relevant knowledge snippets.
4. Context Assembly: Retrieved knowledge is combined with the query.
5. Generation: The assembled context is sent to the LLaMA model to generate a natural, context-rich response.
6. Response Delivery: The chatbot displays the response to the user in an intuitive conversational style.

🧪 Usage

1. Visit Virtual Tuhin Kumar Dutta to interact with the chatbot.
2. Ask any questions related to Tuhin Kumar Dutta’s professional background, expertise, projects, or other related areas.
3. Keep in mind the eco-friendly spirit: minimize unnecessary queries to reduce computational resource consumption.

🔮 Future Enhancements

• Integrate multi-modal inputs (voice, images).
• Expand knowledge base dynamically with new documents.
• Optimize inference speed and reduce latency.
• Add user analytics and feedback collection.

🤝 Contributing

Contributions and suggestions are welcome! Feel free to open issues or submit pull requests. For more details and updates, visit the HuggingFace Spaces Repository.