LLM Prompt Engineering

Getting an LLM to extract recipes consistently is harder than it sounds. Social media recipes are chaos, vague ingredient lists, missing nutrition info, cook times buried in paragraphs, and half the time it's not even a recipe.

I needed the AI to parse "2 tbsp olive oil" and "1 medium onion, diced" into structured data, estimate macros when they're missing, pull cook times from natural language, and skip non-recipe content entirely. All while spitting out clean JSON every time.

The fix: a tightly structured prompt with few-shot examples and hard constraints. Show the model what good output looks like, and it'll follow the pattern.

Now it handles Instagram captions, YouTube descriptions, and recipe websites without breaking format. Prompt engineering is just teaching by example at scale.