TinyWorld Summarization (Offline, Sub-10MB Design)

This project implements a lightweight, fully offline summarization pipeline inspired by the TinyWorld Antigravity spec. It provides a hybrid, routed system with fast-path heuristics, a section-aware extractive summarizer, and a neural-stub interface for future training.

Goals

• Fully offline inference
• CPU-only, low memory
• Modular components
• Summarization focus

Quick Start

python -m examples.demo

CLI:

python -m tinyworld.cli --text "Your text here" --max-sentences 3

Web UI:

python -m tinyworld.webapp

Then open http://localhost:8080 in your browser.

Structure

• tinyworld/input_ingestion.py: validation + normalization
• tinyworld/preprocess.py: language detection (heuristic), tokenization, complexity scoring
• tinyworld/routing.py: routing logic
• tinyworld/fastpath.py: dictionary and template hooks
• tinyworld/heuristic.py: extractive summarizer
• tinyworld/neural_stub.py: placeholder neural interface
• tinyworld/summarizer.py: high-level pipeline
• tinyworld/cli.py: command-line entry
• tinyworld/webapp.py: local web server + API

Notes

• Uses only Python standard library.
• Neural model is a stub; designed for later training and integration.