Find Me a Job: How Search Works on hh.ru
summary
Evolution of hh.ru job search from a simple Lucene-based implementation to a high‑load search cluster with personalization using machine learning, quality metrics and extensive A/B testing. Covers the problems encountered during scaling and the engineering solutions used to deploy and operate this production search system.
AI-written summary · about
chapters (20)
- 0:00hh.ru search intro
- 1:10Search team and scale
- 3:01Search system requirements
- 5:11Empty query problem
- 6:18Sphinx search tech
- 7:47Sphinx customization
- 9:12Inverted index
- 11:40Index sorting
- 13:08Speed and uptime needs
- 15:17Search cluster architecture
- 17:02Timeout handling strategies
- 18:08Ranking principles
- 20:00Personalized search
- 22:30ML model features
- 25:23Model integration in search
- 26:28Model load optimization
- 27:54Quality metrics
- 29:18Experiments and A/B testing
- 31:03Summary and recommendations
- 31:33Q&A
related talks
40mLevel Up: Everything you need to know about search, ...
41mDynamic Ranking of Search and Recommendation Results in High-Traffic Systems
51mHow to Effectively Rank All of Runet's Products
01h 33mEmployer Name Normalization in the Online Recruitment Industry
26mStreamlining Property Search Across Multiple Platforms
38mLearning to Rank: where search meets machine learning #bbuzz
47mHow to deploy a recommendation service with a BERT-like model to high-load production
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz