PromptEngineering
Designing system prompts, few-shot examples, and structured output formats; defending against prompt injection.
Why we need this / what value this brings
A well-designed prompt is the cheapest lever for output quality/safety — far cheaper than fine-tuning or switching models.
When to use this
Any time you're building an LLM feature — write and version the system prompt before writing the surrounding app code.
How to use or implement this
Draft a system prompt with explicit role/constraints/output format, test it against edge cases (empty input, adversarial input), and treat it like code: version it, review changes to it.
Research questions
- What's a minimal, testable system prompt for a support/booking assistant?
- How do you defend a user-facing chat feature against prompt injection from pasted content?
- Structured output: JSON mode / function-calling vs free text parsing — when does it break?
Empty folder — drop notes, links, and findings here as you research.