PaperOrchestra, developed by Google Cloud AI Research, takes unstructured pre-writing materials (such as ideas and experiment logs) as input and outputs a LaTeX manuscript complete with a literature review and figures. The workflow is divided among several agents responsible for outlining, literature review, section-by-section writing, content revision, and plotting. The plotting agent generates charts using code and also supports the insertion of conceptual diagrams. Supported models include Gemini (via Vertex AI or API Key) and OpenAI, while literature retrieval uses the Semantic Scholar API. The project also includes an interactive Streamlit demo and automated evaluation scripts.
Why It Matters Outlining, literature retrieval and review, section-by-section writing, revision, and plotting are assigned to different agents, producing output directly formatted according to the target conference's LaTeX template.
Stars 1541 (+313 / 24h)
Reasons
- 24 hours added 313 stars, total stars 1541, ranked 4 on Trendshift for the day (reflecting attention only); Apache-2.0 license
- Includes an arXiv paper; conference LaTeX templates are located in the templates directory, and users can add new templates
Usage
- Create a Python 3.11 environment and install dependencies, fill in the keys for Semantic Scholar and the chosen model in .env, then run python paper_writing_cli.py --raw_materials_dir materials_directory --latex_template_dir templates/cvpr2025 --use_plotting true
Risks
- The datasets used in the paper's experiments have not yet been released with the repository; the README notes that this is not an officially supported Google product
- Citations and conclusions in the generated draft require line-by-line verification by the author, and you must also check the target conference's policies on AI-assisted writing before submission