Alexandr Yarats, Head of Search at Perplexity AI, discusses his interest in machine learning and his career path trajectory
Perplexity AI is an AI-chatbot-powered research and conversational search engine used to generate answers from web sources
Yarats previously worked at Google Assistant on the Google Bard team as a Senior Software Engineer, before joining Perplexity AI
As the Head of Search at Perplexity, Yarats is responsible for building the internal retrieval system for the search engine
Perplexity shifted from developing a text-to-SQL tool to creating an AI-powered search engine, building on the concept of retrieval-augmented generation (RAG)
Perplexity utilises RAG by having tens of LLMs (ranging from big to small) work together in parallel to handle one user request, improving accuracy
Perplexity focuses on frequently-searched topics when crawling and indexing web pages, using LLMs to maximise signal-to-noise ratio
The search engine ranks objectives designed to natively combine the retrieval system and LLMs, instead of optimising for click probability or ad impression
Perplexity's main technological and market challenges are improving the helpfulness and accuracy of answers, scope and complexity of questions it can answer, and driving costs down