perplexity interview was very search-focused: 1. coding: implement a BM25 scoring function from scratch. then add TF-IDF weighting 2. system design: design a real-time search engine that combines web crawling, indexing, and LLM-powered answer synthesis 3. ML deep dive: discuss the tradeoffs between dense retrieval (embeddings) vs sparse retrieval (BM25) vs hybrid 4. coding: implement a simple RAG pipeline - given a query, retrieve top-k documents and generate a grounded answer they clearly want people who understand both traditional IR and modern LLM-based search

Perplexity senior SWE - search and RAG focused

2297188c

perplexity interview was very search-focused: 1.

0e458127

BM25 from scratch is a great interview question. tests whether you actually understand scoring functions or just import them from a library

34aab1d6

the hybrid retrieval question is the right one to ask. the answer is: sparse for exact keyword matches, dense for semantic similarity, hybrid for production search