Lesson 3 of 3Module 1 · 0% of course
LLMs & AI Engineering · How LLMs work

Prompting & structured output

Get reliable JSON out of an LLM: clear instructions, examples, a schema, and validation.

15 min · 5 steps · videoJump to quiz ↓
How I use LLMs Andrej Karpathy · 131 min
Open in YouTube
Step 1 of 5 · 1. A good prompt has parts15 min read ·

1. A good prompt has parts

Role/system
who it is, rules
→
Context
the data (two trade records)
→
Task
classify the break
→
Format
exact JSON schema
→
Examples
1–3 worked cases
SYSTEM: You classify trade breaks. Answer only JSON.
USER: ours={"id":"T2","qty":250,"px":"101.25"}
      theirs={"id":"T2","qty":200,"px":"101.25"}
Return {"type": "qty|px|missing|duplicate", "reason": str}
Next: 2. Never trust, always validate
‹ PrevNext ›