How nikhilrajput63/cs671-hackthon-project-2025 works
Project type and purpose
• This is a multi-module AI recommendation system that analyzes a user’s mood via three parallel pipelines—facial expressions, text+emoji, and (optional) voice—and suggests movies accordingly.
• Each subdirectory is effectively its own “service” or module with its own models, data processing, and entry-point (app.py).
File structure
• Top-level modules:
– face_expression_recommendation/
– text_emoji_recommendation/
– voice_recommendation/
• Each contains:
– An entry script (app.py)
– Model-loading and inference components (e.g., mood_detector/face_mood.py, models/mood_predictor.py)
– Data processing utilities (data_processing.py, utils/data_processor.py)
– A recommendation engine (recommender/suggestor.py, recommendation_engine.py)
– Static assets or model artifacts (movies.csv, resnet50*.pth)
• The text UI folder follows an MVC-style pattern: ui/components/ for React-like widgets and ui/app.py as the controller.
README insights
• Each pipeline is standalone today; future plans call for unifying all three into one application.
• Dependencies are isolated per directory by requirements.txt.
• Data stores are simple CSV files; no external database yet.
Main components
a. Input sources
– Camera feed (face pipeline)
– Text input & emoji selection (text pipeline)
– Microphone/voice capture (voice pipeline)
b. Pre-processing modules
– face_expression_recommendation/data_processing.py
– text_emoji_recommendation/utils/data_processor.py
– voice_recommendation/voice_processing.py
c. Mood-detection models
– face: mood_detector/face_mood.py + ResNet50 weights
– text: models/text_embedder.py → mood_predictor.py
– voice: (future) voice tone classifier
d. Recommendation engines
– face_expression_recommendation/recommender/suggestor.py
– text_emoji_recommendation/models/recommendation_engine.py
e. Data store
– movies.csv / processed_movies.csv (stored locally)
f. User interface / API
– face: app.py CLI or minimal web interface
– text: ui/app.py + components
– voice: app.py script
Component interactions and data flow
• Model artifact loading from the file system into the detection modules.
• Shared movie database being read by each recommendation engine.
Architectural patterns & principles
• Modular “microservice” style: each mood-detector is self-contained.
• MVC pattern in text pipeline’s UI folder.
• Clear separation of concerns: I/O, preprocessing, inference, recommendation.
• Future orchestration layer (API gateway or aggregator service) can sit in front to choose among pipelines.