The GenAI Field Guide

Interview-ready in an evening

The questions that separate people who have built GenAI systems from people who have only read about them.

1. What is attention?
2. What is RLHF?
3. How do you choose chunk size?
4. What are recall@K and precision@K?
5. Fine-tuning versus RAG?
6. What is LoRA?
7. Can an LLM judge be trusted?
8. What is context rot?
9. Why can leaderboards mislead?
10. What is a KV cache?
11. Why can a better prompt make a system worse?
12. Why isn't more context always better?
13. Why can a small model beat a large model in a system?
14. Why keep critical actions deterministic?
15. Which parts of a system should be probabilistic?
16. How do I debug a bad answer?