How aoi20031020/ripogram-generator works
Project type and purpose
• This is a text-generation tool/library with both Japanese (LLM-based) and English (BERT+WordNet) pipelines.
• It provides two “front ends”: Streamlit web apps (Japanese, English, integrated) and CLI interfaces, plus example scripts.
• Core functionality is packaged under ripogram/ as a reusable library.
File structure
• Top-level directories:
• apps/ – Streamlit UIs
• ripogram/ – core package (CLI modules, config, core rewriting/tokenization logic)
• examples/ – demonstration scripts and a Jupyter notebook
• tests/ – unit tests for both languages
• docs/ – design/docs/PDF slides
• Key config files: .env.example, requirements.txt
• No persistent database; external dependencies include OpenAI API (GPT-4), SpaCy, WordNet.
README for additional insights
• Japanese pipeline uses an LLM via OpenAI API.
• English pipeline uses SpaCy+pretrained BERT embeddings + WordNet synonym replacement.
• Configurable via ripogram/config.py pulling from environment (.env).
• Users launch with Streamlit or CLI (ripogram.cli, ripogram.english_cli).
Detailed separation of components
• UI Layer:
• Streamlit Apps
• CLI modules
• Application Orchestrator: entry scripts → config loader
• Core Library Subsystems:
• Tokenization → Rewriting → Utilities
• External Services:
• OpenAI GPT-4
• SpaCy en_core_web_sm
• WordNet database
• Demonstration/Test:
• examples/
• tests/