Foundation Model Prompting — Maximise Model Performance

I will optimise your prompting approach for your chosen foundation model — testing system prompt variations, few-shot example selection strategies, chain-of-thought reasoning elicitation, output format instructions, and temperature/top-p sampling parameters — to maximise your model's performance on your specific task without fine-tuning. Foundation model prompting is more model-specific than most practitioners realise — the prompting techniques that maximise GPT-4o performance often differ substantially from those that maximise Claude or Gemini performance on the same task, and systematic optimisation using your actual evaluation data consistently outperforms intuitive prompt design.

The optimisation covers a systematic prompt experimentation framework, system prompt variant testing, few-shot example curation and testing, chain-of-thought elicitation for reasoning tasks, output format specification testing, sampling parameter optimisation, and a final optimised prompt set with evaluation evidence.