LangChain-based-psychological-wellness-RAG-bot

🤖 AI-Powered Multimodal Chatbot with Document Q&A and Movie Recommender System

This project is a voice + text enabled conversational assistant that combines the power of Large Language Models (LLMs), document understanding, and content-based movie recommendations into a single Streamlit web application.

It is built with:

• 🧠 Mistral-7B-Instruct for natural language responses
• 📄 FAISS + HuggingFace Embeddings for vectorstore and semantic search over documents respectively
• 🎬 TMDB API for movie retrieval based on keywords.
• 🎤 Whisper for voice input transcription
• 💻 Streamlit for a clean, user-friendly interface

---

Tech Stack

Programming Languages

• Python

Libraries & Frameworks

• Langchain: For handling natural language processing (NLP) tasks and integrating LLMs.
• Hugging Face: For using the Mistral-7B-Instruct model and embeddings.
• FAISS: For vector storage and retrieval.
• Streamlit: For building the web interface and interactive features.
• Whisper: For speech-to-text transcription.
• Sounddevice: For recording audio from the microphone.
• Pandas: For data manipulation and handling movie data.
• Requests: For interacting with external APIs.
• PyPDFLoader and DirectoryLoader: For loading and processing PDF documents.

Models

• Mistral-7B-Instruct: A large language model from Hugging Face used for text generation and answering questions.
• Whisper: A speech-to-text model used for transcribing audio input.

APIs

• The Movie Database (TMDB) API: Used for fetching movie posters based on movie IDs.

Data Storage

• FAISS: For storing and retrieving embeddings of documents and movie information.

---

📙Dataset Used

• The GALE ENCYCLOPEDIA of MEDICINE SECOND EDITION
• SELF MADE EMOTIONAL SUPPORT DATASET

---

🧩 Project Structure

📍 Phase 1: Creating memory for LLM

🔹 Objective:

Enable the chatbot to answer questions based on the content of uploaded documents (PDFs).

🔹 Implementation:

• Document Loading: PDFs are loaded using PyPDFLoader.
• Text Splitting: Documents are split into overlapping chunks using CharacterTextSplitter for better semantic continuity.
• Embedding Generation: Each chunk is embedded using HuggingFace’s all-MiniLM-L6-v2.
• Vector Indexing: FAISS is used to index and store the embeddings for efficient retrieval.

🔹 Prompt Template:

Ensures that answers remain based on document context and avoids hallucination by instructing the model to say “I don't know” if unsure.

---

📍 Phase 2: Connecting Memory with LLM

🔹 Objective:

Allows to connect LLM with FAISS and create QA chain

🔹 Implementation:

• Setup LLM: Setting LLM (Mistral with HuggingFace)
• Connect LLM+ VectorStore: Connection of LLM and FAISS for data retrievel.
• Loading Database: Loading database for vectorstore directory.
• QA Chain creation

---

📍 Phase 2: Voice-to-Text Interaction

🔹 Objective:

Allow users to interact with the chatbot via voice in addition to text.

🔹 Implementation:

• Records audio using sounddevice.
• Transcribes audio using OpenAI’s Whisper model.
• Transcribed text is automatically sent to the chatbot for processing.

---

📍 Phase 4: Streamlit UI & Styling

🔹 Objective:

Deliver a clean, user-friendly interface with multi-modal input support and added user engagement.

🔹 Implementation:

• Chat Interface: Supports both text and voice input, and retains multi-turn history via st.session_state.
• File Uploader: For uploading PDFs used in document Q&A.
• Custom Styling: Background and input fields styled using inline CSS.
• Dynamic Outputs: Chat responses, transcribed audio, and movie posters rendered in real-time.
• Integrated Movie Recommendation System: Seamlessly embedded the movie recommendation model into the interface. If a user mentions feeling unwell or needs a break, the chatbot suggests relevant movies based on keywords or Title — enhancing user satisfaction and engagement.

---

📸 Project Screenshots

🔹 Streamlit Chatbot Interface

🚀 Getting Started

🛠️ Prerequisites

• Python 3.9+
• Install dependencies:

pip install -r requirements.txt