Cine-RAG is an advanced Retrieval-Augmented Generation (RAG) system that revolutionizes movie recommendations by understanding semantic intent rather than relying on simple keyword matching. Unlike traditional recommendation systems that depend on user viewing history, Cine-RAG leverages a Vector Database to comprehend natural language queries such as:
- "Dystopian movies where teenagers fight to the death"
- "Action movies released after 2022 with high ratings"
- "Psychological thrillers with unreliable narrators"
- Utilizes
all-MiniLM-L6-v2embeddings to capture the "essence" of your request beyond simple keyword matching - Understands context, themes, and narrative elements in natural language
- Combines vector similarity search with metadata filtering
- Support for complex queries like:
Year > 2023 AND Genre = Action AND Rating > 7.5
- Runs 4-bit quantized Zephyr-7B-Beta locally on Tesla T4 GPU
- No data sent to external APIs
- Completely free and privacy-preserving
- Implements strict "Evidence-Based Reasoning" prompts
- Prevents model from fabricating non-existent movies
- Chain-of-Thought prompting ensures factual accuracy
- Uses Maximum Marginal Relevance (MMR) algorithm
- Ensures diverse, non-redundant recommendations
- Balances relevance with variety
| Component | Technology |
|---|---|
| Orchestration | LangChain |
| Vector Database | ChromaDB |
| LLM | HuggingFace Zephyr-7B-Beta (4-bit quantized) |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Data Processing | Pandas (1M+ rows → 27K curated entries) |
| Frontend | Gradio |
| Quantization | bitsandbytes |
graph LR
A[User Query] --> B[Query Processing]
B --> C[Filter Detection]
B --> D[Embedding Generation]
C --> E[ChromaDB Vector Search]
D --> E
E --> F[MMR Retrieval]
F --> G[Top 4 Relevant Movies]
G --> H[Zephyr-7B LLM]
H --> I[Grounded Recommendation]
-
Data Ingestion
- Loads curated dataset of 27,000+ movies with metadata and plot summaries
- Each movie includes: title, genre, year, rating, plot, and more
-
Vector Embedding
- Text descriptions converted into 384-dimensional vectors
- Stored in ChromaDB for lightning-fast similarity search
-
Intelligent Retrieval
- Detects temporal filters (e.g., "after 2020")
- Extracts genre preferences
- Retrieves top 4 most relevant movies using MMR
-
Grounded Generation
- Retrieved data passed to Zephyr-7B with strict grounding prompt
- LLM generates recommendations with explicit reasoning
- Citations to prevent hallucinations
You can run this project completely free on a Tesla T4 GPU using Kaggle Kernels. open Notebook
Prerequisites
- Python 3.10+
- NVIDIA GPU with 12GB+ VRAM (for local deployment)
- CUDA 11.8+ and cuDNN
- Clone the Repository
git clone https://github.com/Mohammed2372/Movies-Recommendation-Chatbot-RAG.git
cd Movies-Recommendation-Chatbot-RAG- Install Dependencies
pip install -r requirements.txt- Run Chatbot
Run all cells in chatbot.ipynb notebook. A frontend UI is available for interactive testing.
Problem: The 7B model would fabricate details, claiming a 2023 movie was released in 2021 or inventing non-existent films.
Solution: Implemented Chain-of-Thought (CoT) prompting that forces the model to explicitly state evidence before making claims.
Problem: Raw TMDB dataset contained 1M+ entries, including unreleased clips, duplicate entries, and low-quality productions.
Solution: Engineered robust Pandas pipeline with filters (vote_count > 50, status == 'Released', deduplication) resulting in 27,000 high-quality movies.
Problem: Vector search would return very similar movies (e.g., all Marvel films for "superhero" query).
Solution: Implemented Maximum Marginal Relevance (MMR) to balance relevance with diversity across different sub-genres.
User: "Movies about artificial intelligence gaining consciousness"
Response:
Based on your query, here are recommendations:
1. **Ex Machina** (2014) - Drama/Sci-Fi
A programmer is selected to evaluate the human qualities of a breathtaking AI.
2. **Her** (2013) - Romance/Sci-Fi
A lonely writer develops an unlikely relationship with an operating system...
User: "Action movies from 2023 with ratings above 8"
Response:
Here are high-rated action films from 2023:
1. **John Wick: Chapter 4** (2023) - Action/Thriller [8.2/10]
John Wick uncovers a path to defeating The High Table...
- TMDB for the movie dataset
- Hugging Face for Zephyr-7B and transformers library
- LangChain team for the RAG framework
- ChromaDB for the vector database
- Kaggle for free GPU resources and datasets
- Datasets