CrackML by @ml.with.umang
GenAI interview guides / RAG System Design Interview Guide
GenAI / LLM

RAG System Design Interview Guide

Design retrieval-augmented generation systems that are measurable, secure, low-latency, and robust to bad retrieval.

Start with the product contract

Define freshness, latency, citation, permissions, abstention, and failure costs before choosing embeddings or a vector database.

Retrieval pipeline

Separate ingestion, parsing, chunking, metadata, embedding, candidate retrieval, reranking, context assembly, generation, and citation validation.

Evaluation

Measure retrieval quality separately from answer quality. Include hard negatives, stale docs, permission-sensitive cases, and unanswerable questions.

Production trade-offs

Discuss chunking, top-k, reranking depth, caching, freshness, tenant isolation, fallbacks, latency, and cost.