🤖 Chatbot Q&A

It is a local Retrieval-Augmented Generation (RAG) chatbot built with LangChain, FAISS, and Streamlit. It uses an eCommerce FAQ dataset to answer user queries by retrieving semantically similar Q&A pairs and generating human-like responses using a locally hosted Qwen2 LLM via Ollama. Embeddings are created using the hkunlp/instructor-large model from HuggingFace. The app includes a built-in UI for interacting with the chatbot and regenerating the knowledge base on demand.

---

🚀 Overview

chatbot-q-and-a is an intelligent chatbot application built to answer common questions from an eCommerce FAQ dataset. It utilizes powerful LLMs and vector-based semantic search to deliver accurate and summarized responses. If the bot doesn't know the answer, it politely says so — no hallucinations here!

This app supports regenerating embeddings through a button on the UI, offering flexibility to refresh or update the knowledge base with ease.

---

🧠 Features

• ⚡ Embedding-based question retrieval with FAISS
• 📚 Dataset: ecommerce_FAQ_chatbot_dataset on Kaggle
• 🔍 Uses hkunlp/instructor-large from HuggingFace
• 🧠 Summarizes multiple similar answers using an LLM
• 💬 Natural UI using Streamlit
• 🛠️ Local inference using Ollama + Qwen2 LLM
• 🔁 One-click embedding regeneration from UI
• ❌ Gracefully handles out-of-scope questions by responding with "I don't know."

---

🛠️ Tech Stack

Tool / Package	Purpose
streamlit	Frontend web UI
langchain	LLM & prompt orchestration
langchain-community	Integrations with external tools
faiss-cpu	Fast similarity search on CPU
tiktoken	Tokenization support
protobuf	Serialization (LangChain & Ollama)
sentence-transformers	Transformer embedding utilities
InstructorEmbedding	HuggingFace embedding wrapper
jq	JSON parsing (if used in post-processing)

---

📂 Dataset

We are using the ecommerce_FAQ_chatbot_dataset, which includes real-world Q&A pairs commonly asked by customers.

---

🔧 How It Works

1. Embedding Creation

   • The Q&A pairs are embedded using the hkunlp/instructor-large model from HuggingFace.
   • These embeddings are stored using FAISS for efficient similarity search.

2. User Question Flow

   • User asks a question via the Streamlit interface.
   • The question is converted to an embedding using the same model.
   • FAISS retrieves the closest match(es) from the dataset.
   • The matched answer(s) are fed to the LLM (Qwen2 via Ollama).
   • If multiple matches are returned, the LLM summarizes them.
   • If no good match is found, the LLM responds: "I don't know."

3. Knowledge Base Regeneration

   • A dedicated UI button allows regenerating embeddings from scratch.

🚀 Getting Started

1️⃣ Install Ollama and Pull the Llama3 Model

Make sure you have Ollama installed.

Then pull the qwen2 model locally:

ollama pull qwen2

2️⃣ Clone the Repository

git clone https://github.com/js-kalsi/restaurent-genai.git
cd restaurent-genai

3️⃣ Set Up a Virtual Environment & Install Dependencies

We recommend using a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Then install dependencies:

pip install -r requirements.txt

4️⃣ Run the App

Launch the Streamlit app:

streamlit run app.py

The app will open in your default web browser. Select a cuisine, and let the AI generate a restaurant concept for you!