How kartikeya-datta/tv-series-analysis-system works
Project type and purpose
• This is a full-stack, Python-based ML web application for “Naruto” subtitle analysis. It provides four core features—theme analysis, named-entity recognition + character network, jutsu text classification, and a Naruto character chatbot—all exposed via a Gradio UI.
File structure
• Top level
• gradio_app.py – entrypoint for the web interface (Presentation + Controller)
• requirements.txt, .gitignore, .env_example – config & dependencies
• data/ – raw subtitles, jutsu JSON, CSVs, images
• stubs/ – sample outputs for fast front-end rendering
• notebooks (.ipynb) under each module for experimentation and model training
• Feature modules (each acts like its own service within the monolith)
• theme_classifier/ – zero-shot theme classification using Hugging Face BART
• text_classification/ – jutsu classifier training & inference with DistilBERT
• character_network/ – spaCy NER + NetworkX/PyVis graph generation
• crawler/ – web crawler for jutsu data collection
• utils/ – shared data loaders and training utilities
README for additional insights
• It describes four feature pipelines (theme, NER/network, jutsu, chatbot), their data-processing steps, and the tech stack (Gradio, Hugging Face Transformers, spaCy, NetworkX, PyVis, scikit-learn, LoRA, 4-bit quantization, etc.).
• It mentions model training flows in notebooks and inference flows in the Gradio app.