ai-ml pattern
RAG Systems
Retrieval-Augmented Generation - ground LLM responses in your own data. Essential for enterprise AI.
Time
O(log n) retrieval with vector DB
Space
O(n) for embeddings storage
🧠Mental Model
“An open-book exam - the LLM can look up answers in your documents before responding.”
Verbal cue: Retrieve relevant context, then generate informed answers.
🎯Recognition Triggers
When you see these patterns in a problem, consider this approach:
RAGknowledge basechat with documentsenterprise AIgrounded responses
💡Interview Tips
- 1Know trade-offs: chunk size, overlap, embedding model
- 2Mention hybrid search: vector + keyword
- 3Discuss evaluation: retrieval accuracy, answer quality
⚠️Common Mistakes
- ✕Chunks too large (bad retrieval) or too small (lost context)
- ✕Not handling "no relevant context" cases
- ✕Ignoring metadata for filtering