Read the research,
fork the code
Bad Metaphor emerges from research on voice-driven AI pedagogy and the failure of rubric-based creative evaluation. The full system — training pipeline, model weights, and orchestration code — is open source.
📄
Voice-Driven Poetry Pedagogy: Teaching Machines to Teach Verse Through Aesthetic Conviction
Our research paper on training educator models with genuine voice rather than neutral evaluation. Introduces the cliché autopsy methodology and measures pedagogical effectiveness against rubric-based baselines.
arxiv.org/abs/XXXX.XXXXX →⌨bad-metaphor — Full Source Repository
Training pipeline (Modal + Claude API data generation), QLoRA fine-tuning scripts, GGUF quantization, multi-agent orchestration, and the Next.js workshop interface. MIT licensed.
github.com/your-username/bad-metaphor →