No Priors Ep. 8 | With Neeva’s Sridhar Ramaswamy

For the first time in decades web search might be at risk for disruption. Bing is allied with OpenAI to integrate LLMs. Google has committed to launching new products. New startups are emerging. Sridhar Ramaswamy co-founded the challenger AI-powered, private search platform Neeva in 2019. He is a former 16-year Google veteran who most recently led the internet’s most profitable business as SVP in charge of Google Ads, Commerce and Privacy. Sridhar, Elad and Sarah talk about the challenge of building search, how LLMs have changed the landscape, and how chatbots and "answer services" will affect web publishers. 00:00 - Introduction 01:32 - Why Sridhar started a private search engine after leaving Google 11:11 - Information Retrieval Problems, Mapping Search Queries and LLMs 15:25 - Google and Bing’s approach to search with LLMs 19:06 - Scale challenges when building a search engine startup 22:26 - Distribution challenges and why they release Neeva Gist 24:11 - Why Neeva is a privacy centric subscription service 28:25 - The relationship between search and publishers/content creators 30:16 - Sridhar’s predictions on how AI will disrupt current ecosystems

Sarah GuohostSridhar RamaswamyguestElad Gilhost

Apr 24, 202335mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

Neeva’s Sridhar Ramaswamy on Reinventing Search With AI Answers

Sridhar Ramaswamy discusses why he left Google to build Neeva, a privacy-focused, ad-free search engine that later pivoted to AI-powered answer experiences. He explains how large language models (LLMs) unlocked scalable, cited summaries that fundamentally improve search versus traditional lists of links. The conversation dives into the technical and economic challenges of marrying LLMs with information retrieval, from cost and infrastructure to model selection and retrieval-augmented generation. They also explore distribution hurdles, business models, and how AI answers may reshape the relationship between search engines, advertisers, and content publishers.

IDEAS WORTH REMEMBERING

5 ideas

Ad-free, privacy-first search alone isn’t enough to shift consumer habits.

Neeva learned that while people say they care about privacy and fewer ads, especially in the US they rarely switch defaults unless there is a clear, tangible experience upgrade; Europe showed more sensitivity to privacy.

AI-generated, cited answers are a step-function improvement over link lists.

LLMs now let search engines synthesize authoritative summaries for the majority of answerable queries, aligning with users’ core preference for minimal effort and direct answers rather than scanning opaque links.

LLMs can replace many bespoke, brittle systems in search infrastructure.

Tasks like query rewriting, structured data extraction, and trigger logic for rich results that once required complex regexes and custom code can now be handled by small, fine-tuned models, dramatically simplifying engineering.

Cost challenges of LLM search can be mitigated with smaller, specialized models.

Instead of relying on a single massive model for everything, Neeva uses 5–10B parameter models, fine-tuning and human feedback to solve specific problems at much lower serving costs while preserving quality.

Distribution and default habits remain the existential challenge for new search engines.

Even with better technology, dislodging entrenched defaults like Google requires creative distribution strategies—apps like Gist, integrations with products like Dashlane or Proton, and embedded conversational search for publishers.

WORDS WORTH SAVING

5 quotes

It cannot be that there's one company, one religion, one product for the whole world.

— Sridhar Ramaswamy

If you can provide a believable answer to a question, people are always going to prefer that over any number of links.

— Sridhar Ramaswamy

Search is one of these things where you need a fair amount of scale before you have any kind of meaningful product.

— Sridhar Ramaswamy

Smart people will come up with great explanations for everything that they do as long as it's convenient to them.

— Sridhar Ramaswamy

If two years from now three college kids can build a brand new application that uses language models in a fundamental way, that feels very possible—and that is what is really exciting.

— Sridhar Ramaswamy

Motivation for founding Neeva and limitations of incumbent searchPrivacy, ad-free search, and consumer behavior differences (US vs. Europe)AI summaries, cited answers, and retrieval-augmented generation in searchCost, scale, and model-size tradeoffs for LLM-powered searchTechnical and organizational challenges of building a new search engineDistribution, partnerships, and business model choices (subscriptions vs. ads)Long-term impact of AI on publishers, content creation, and AI-driven tools

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.