LLM Mobile Movie Search

This is a full-stack mobile application that leverages LLMs and vector embeddings to provide a semantic search and conversational recommendation engine for movies.

movie_recommendation_llm.mp4

🌟 Overview

The project consists of a React Native (Expo) frontend and a Python (FastAPI) backend. It offers two primary ways for users to discover movies:

1. Semantic Search: A traditional search screen where users can type in a description, a theme, or a vague idea of a movie, and the system will find the most relevant matches based on semantic meaning, not just keywords.
2. Conversational Chat: A chat interface where users can interact with an AI assistant. Users can ask for recommendations, refine their search, and get movie suggestions in a natural, conversational flow.

🛠️ Tech Stack

Area	Technology
Backend	Python, FastAPI, Sentence-Transformers (for embeddings), ChromaDB/FAISS (for vector search)
Frontend	React Native, Expo, TypeScript, Expo Router
Data	The Movie Database (TMDB) API
AI/LLM	OpenAI API (or similar, for the chat functionality)

🚀 Getting Started

Prerequisites

• Node.js and npm/yarn
• Python 3.8+ and pip
• A TMDB API Key
• An OpenAI API Key (or another LLM provider)
• Expo Go app on your mobile device for testing.

Backend Setup

1. Navigate to the backend directory:

   cd backend

2. Install Python dependencies:

   pip install -r requirements.txt

3. Set up environment variables: Create a .env file in the backend directory and add your API keys:

   TMDB_API_KEY="your_tmdb_api_key_here"
   OPENAI_API_KEY="your_openai_api_key_here"

4. Fetch and Process Data: Run the script to download movie data from TMDB. This will create a movies_with_genres.csv file in the data/ directory.

   python src/database/fetch_data_from_tmdb.py

   (Note: A second script to generate and store embeddings in a vector database would also be run here, e.g., python src/database/generate_embeddings.py)

5. Run the backend server:

   uvicorn src.main:app --host 0.0.0.0 --port 8000 --reload

   The server will be available at http://<YOUR_LOCAL_IP>:8000.

Frontend Setup

1. Navigate to the frontend directory:

   cd frontend

2. Install Node.js dependencies:

   npm install
   # or
   yarn install

3. Configure the API URL: Open the following files and update the API_BASE_URL constant to point to your backend server's local IP address:

   • frontend/app/(tabs)/index.tsx
   • frontend/app/(tabs)/chat.tsx

   // Replace <YOUR_MACHINE_LOCAL_IP> with your actual IP
   const API_BASE_URL = "http://<YOUR_MACHINE_LOCAL_IP>:8000/api";

4. Start the development server:

   npx expo start

5. Scan the QR code with the Expo Go app on your iOS or Android device to run the application. Ensure your device is on the same Wi-Fi network as your computer.

⚙️ How It Works

1. Data Ingestion & Embedding

• The fetch_data_from_tmdb.py script queries the TMDB API for a list of popular movies.
• It saves relevant data (title, overview, genres, etc.) into a CSV file.
• A separate process (not shown in context) would then read this CSV, use a model like Sentence-Transformers to convert the movie overviews into numerical vector embeddings, and store these embeddings in a vector database like ChromaDB.

2. Semantic Search (/api/search)

1. The React Native app sends the user's query text to the FastAPI backend.
2. The backend converts the incoming query into an embedding using the same sentence-transformer model.
3. It performs a similarity search (e.g., cosine similarity) in the vector database to find the movie embeddings that are "closest" to the query embedding.
4. The top N results are returned to the frontend and displayed in a list.

3. Conversational Search (/api/chat_search)

1. The user sends a message in the chat UI.
2. The frontend sends the entire conversation history to the /api/chat_search endpoint.
3. The backend forwards this conversation to an LLM (like GPT-4). The LLM is prompted to act as a helpful movie assistant and is given access to a "tool" which is our semantic search function.
4. If the user's request requires finding movies, the LLM calls the search tool with an appropriate query.
5. The backend executes the search, gets the movie results, and passes them back to the LLM.
6. The LLM formats a natural language response, which might include a sentence like "Here are some movies you might like," and includes the structured movie data.
7. The backend sends this complete response to the frontend, which then renders the text and the movie recommendation carousel.