Pseudowrite
A demonstration tool built for an Augsburg MFA panel on AI and writing. It takes rough story ideas and works them through a three-stage pipeline into structured act outlines and chapter-level notes, using iterative AI refinement loops.
How it works
Stage 1 — Idea to acts: Your rough premise gets expanded into a three-act structure.
Stage 2 — Acts to chapters: Each act gets broken down into chapter-level beats.
Stage 3 — Refinement: The system loops back through what it’s produced, checking and tightening the logic.
The approach uses Chain-of-Thought prompting to work within LLM constraints and get coherent, structured output rather than sprawling prose. There’s also a Playground mode for direct model interaction.
Powered by GPT-3.5 and GPT-4.
What this is and isn’t
Pseudowrite generates structure - act breakdowns, chapter notes, beat sheets. It’s most useful for formulaic genres where structure is well-defined. It won’t write your novel for you, and it’ll produce garbage often.
Not for commercial use. The README is explicit about this, and the ethical concerns around AI training data are real.
How to use
- Download
Pseudowrite_public.ipynb - Add your OpenAI API key
- Run via Jupyter Notebook or Google Colab