A serverless movie recommendation system built on AWS, featuring semantic search with ONNX models, collaborative filtering, and comprehensive user management capabilities.

• Semantic Search: Natural language movie search using ONNX-optimized embeddings stored in S3
• Collaborative Filtering: Personalized recommendations based on user ratings and behavior
• Content-Based Filtering: Recommendations based on movie similarity using weighted embeddings
• User Management: JWT authentication, favorites, reviews, and activity tracking
• Scalable Architecture: Single Lambda function with centralized routing and configuration
• ONNX Model Support: Optimized inference without heavy ML dependencies

The system uses a single Lambda function architecture with:

• Single Entry Point: lambda_handler.py routes all API requests
• Centralized Configuration: utils/config.py manages all environment variables
• DynamoDB: 5 tables for users, movies, reviews, favorites, and activity
• S3 Storage: Movie embeddings (.npz format) and ONNX models
• HTTP API Gateway: AWS HTTP API (not REST) for endpoint management
• Lambda Layers: Optimized dependency management

• Layer 1: PyJWT==2.8.0, bcrypt==4.1.2, onnxruntime, tokenizers
• Layer 2: numpy<1.27.0

1. Setup AWS Infrastructure

   cd initial_setup
   python create_table.py

2. Deploy Single Lambda Function

   REM Package entire application
   powershell Compress-Archive -Path lambda_handler.py,lambda_functions,utils,requirements.txt -DestinationPath movie-recommender.zip
   aws lambda create-function --function-name Movie_recommender_system --runtime python3.9 --zip-file fileb://movie-recommender.zip

3. Configure Environment Variables

   set JWT_SECRET=your-secret-key
   set EMBEDDINGS_BUCKET=movieembeddings
   set MODEL_BUCKET=movieembeddings

4. Setup API Gateway

   cd initial_setup
   python api_gateway_setup.py

All endpoints are routed through a single Lambda function via HTTP API Gateway:

• POST /auth/register - User registration with email/password
• POST /auth/login - User login returning JWT token
• POST /auth/refresh - Refresh JWT token

• POST /search - Semantic search using ONNX model inference
• POST /content - Content-based recommendations from movie IDs with ratings
• POST /collaborative - Collaborative filtering (requires authentication)
• POST /similar - Find similar movies to a given movie

• GET /user-data/favorites - Get user's favorite movies
• POST /user-data/favorites - Add movie to favorites
• DELETE /user-data/favorites/{movieId} - Remove from favorites
• GET /user-data/favorites/toggle/{movieId} - Check if movie is favorited
• POST /user-data/reviews - Add movie review with rating
• GET /user-data/reviews - Get user's reviews
• DELETE /user-data/reviews/{movieId} - Remove review
• GET /user-data/reviews/toggle/{movieId} - Check if movie is reviewed
• GET /user/activity - Get user activity history
• DELETE /user/account - Delete user account

• Format: .npz (NumPy compressed archive)
• Structure: 385 columns (384 float embeddings + 1 string movie_id)
• Storage: S3 bucket movieembeddings

• Model Files: config.json, model.onnx, tokenizer.json, vocab.txt
• Storage: S3 bucket movieembeddings/model_onnx/
• Purpose: Optimized inference without sentence-transformers dependency

1. Movies - Movie metadata (title, overview, genres, etc.)
2. Reviews - User ratings and reviews (user_id, movie_id, rating, timestamp)
3. MovieRecommender_Users - User authentication (email, username, password_hash, salt)
4. MovieRecommender_Favorites - User favorites (user_id, movie_id, created_at)
5. MovieRecommender_Activity - User activity logs (user_id, timestamp, action, data)## Development

For detailed setup instructions, see:

• DEPLOYMENT_GUIDE.md - Complete deployment walkthrough
• API Reference - OpenAPI 3.0 specification
• System Architecture - Detailed component overview

All environment variables are managed centrally in utils/config.py:

# Required
JWT_SECRET = "your-jwt-secret"
EMBEDDINGS_BUCKET = "movieembeddings"

# Optional (with defaults)
JWT_EXPIRY = 7200  # 2 hours
MAX_RESULTS = 100
DEFAULT_TOP_K = 10

Download the following CSV files from Kaggle and place them in the initial_setup directory:

• movies_metadata.csv
• credits.csv
• ratings_small.csv (or ratings.csv for full dataset)

Kaggle dataset: https://www.kaggle.com/datasets/rounakbanik/the-movies-dataset

The system uses Lambda Layers for optimized dependency management:

PyJWT==2.8.0
bcrypt==4.1.2
onnxruntime
tokenizers

numpy<1.27.0

PyJWT==2.8.0
bcrypt==4.1.2
numpy<1.27.0
onnxruntime
tokenizers

All the endopoints are tested on AWS lambda using JSON files in the tests directory. Each test file corresponds to a specific endpoint and includes sample requests and expected responses.

They are also tested using curl commands for remote testing. For example:

curl -X POST http://localhost:3000/auth/login -H "Content-Type: application/json" -d '{"email": "[email protected]", "password": "password"}'
curl -X POST http://localhost:3000/search -H "Content-Type: application/json" -d '{"query": "An action movie with a female superhero"}'
curl -X GET http://localhost:3000/user-data/favorites -H "Authorization: Bearer <token>"

MIT License