Privacy-Preserving Web Search

on YouTube ↗18m

" summary

Quant operates a privacy-first search stack: no tracking cookies and only hashed IPs, no sessions, clicks logged but de-identified. Traffic is large — about 2.2 billion monthly requests (≈6M users) and ~200M queries/month with ~70% unique. Built a custom crawler running 3k URLs/sec (goal 10k), indexes ~5 billion indexed pages on Vespa, and uses a two-stage learned ranker (linear + LightGBM ranking), with Cascade click model to de-bias isolated clicks. No vector search in production; monetization is contextual ads.

AI-written summary · about

" chapters (7)

1. 0:00Introduction to Quant
2. 1:05Company and search context
3. 3:15Privacy-focused search
4. 6:42Ranking and crawling pipeline
5. 10:34Improving relevance
6. 14:05Outlook and future goals
7. 15:15Q&A session

" related talks

• 40mPrivate Web Search with TiptoeBurt Kaliski, Alexandra Henzinger · RSA Conference 2024
• 36mThe Forbidden Index: How We Turned OpenSearch Into a Privacy-Preserving SuperheroU. Mishra, A. Khanna · OpenSearchCon Europe 2026
• 21mBuilding Privacy-First Search With Containerized OpenSearch and Local LLMsR. Karpe, S. Soni · OpenSearchCon Korea 2025
• 42mGoodbye Tracking, Hello PrivacyNina Müller, Lara Menéndez García · Berlin Buzzwords 2022
• 17mPublishing Wikipedia Project Usage Data with Strong Privacy Protections and w/o TrackingHal Triedman · PEPR '22
• 26mThe Future of SearchBrendan Eich, Marc Al-Hames, Sridhar Ramaswamy · DLD Munich 23
• 49mBeyond Cookies: The Unseen Privacy Risks of Web AnalyticsMitchell Green, Heidi Wachs · RSA Conference 2024
• 32mBLENDER: Enabling Local Search with a Hybrid Differential Privacy ModelBrendan Avent, Aleksandra Korolova, David Zeber, Torgeir Hovden, Benjamin Livshits · 26th USENIX Security Symposium (USENIX Security '17)