|
| 1 | +# Changelog |
| 2 | + |
| 3 | +All notable changes to this project will be documented in this file. |
| 4 | + |
| 5 | +The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), |
| 6 | +and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). |
| 7 | + |
| 8 | +## [1.0.0] - 2026-01-24 |
| 9 | + |
| 10 | +### Added |
| 11 | +- Initial release of Crisis Response Data Pipeline |
| 12 | +- Synthetic crisis scenario generation with LLM support (OpenAI, Anthropic, Google Gemini) |
| 13 | +- Multi-perspective response generation (civilian and first responder roles) |
| 14 | +- Structured data output with facts, uncertainties, analysis, and guidance |
| 15 | +- Support for 40+ crisis categories based on FEMA, WHO, UNDRR, and Red Cross classifications |
| 16 | +- Three quality assurance levels |
| 17 | +- Parallel processing for faster generation |
| 18 | +- Progress tracking with ETA |
| 19 | +- Resume capability for interrupted generation |
| 20 | +- Training format conversion (instruction, conversational, completion) |
| 21 | +- Comprehensive CLI interface |
| 22 | +- Unit tests with Python's built-in unittest |
| 23 | +- Complete documentation suite |
| 24 | +- Hugging Face dataset preparation with 2000 training examples |
| 25 | + |
| 26 | +### Features |
| 27 | +- **Multi-Provider LLM Support**: OpenAI GPT-4o-mini, Anthropic Claude-3-5-Haiku, Google Gemini-2.0-Flash |
| 28 | +- **Optimized Cost**: Default configuration costs ~$11.18 for 2000 samples |
| 29 | +- **High Performance**: ~9-13 seconds per sample, ~250-400 samples/hour |
| 30 | +- **Production Ready**: Automatic retry logic, error handling, validation |
| 31 | +- **Training Ready**: Direct conversion to fine-tuning formats |
| 32 | + |
| 33 | +### Technical Details |
| 34 | +- Python 3.13+ compatible |
| 35 | +- Uses Pydantic for data validation |
| 36 | +- LangChain for prompt management |
| 37 | +- JSONL output format |
| 38 | +- MIT License with attribution requirement |
0 commit comments