I've been kicking around an idea and want to read up before I build it.
The idea: you type a rough, half-formed prompt, an LLM works out what you actually mean, and then rewrites it into a specific, optimized prompt tuned for each target model (Claude, GPT, Gemini, etc.).
Before I start building, I'd love to find prior work. Specifically, I'm looking for:
Is this already well-studied and I'm just searching the wrong terms?
If you've run across papers, GitHub repositories, or blog posts tackling model-specific prompt adaptation, please drop them in the comments below! What keywords should I actually be searching for?
source & further reading
dev.to — original article
I Couldn’t Fix My LLM Costs Until I Measured Tokens Per Feature
Small Model SWE‑bench: What Happens When You Push Tiny Models Into Full Task Pipelines
Grok 4.5 Isn't Open Source. The Apache 2.0 Release Has a Privacy Catch.