AI-Assisted Writing Workflow
Local Qwen Chinese Fiction Pipeline
A Python workflow for Chinese fiction generation with Qwen. It separates planning from drafting and passes character notes, prior events, and recent text into the next segment. Saved progress and lightweight text checks make it easier to review or resume a draft.
Period
May - Aug 2026
Tools / Tech
PythonQwenLocal LLMPrompt WorkflowContext Management
Why I built it
I used the project to explore prompt sequencing, lightweight memory, and long-form generation control with a local model.
Links
What it includes
- Runs Qwen through a Python-controlled generation loop, usually with local Ollama, with an optional compatible API provider.
- Builds a fiction workflow for setting, outline, segmented continuation, context review, and draft output.
- Uses a lightweight memory mechanism so previous questions, answers, and story context can guide later writing, with manual review for coherence.
What I worked on
- Designed staged planning and continuation prompts, including self-questioning steps and explicit context handoff.
- Used AI assistance to implement prompt sequencing, saved story state, and pause/resume behavior in Python.
- Reviewed drafts for repeated scenes, inconsistent character details, and abrupt transitions.
What I Learned
- Learned how fragile long-form generation becomes when context is not explicitly managed.
- Practiced designing memory and prompt steps as a workflow instead of a single prompt.
- Built intuition for the gap between an agent-like effect and a reliable agent system.