Product Manager · Rambler&Co Search
Local Search for Rambler&Co
The search team's next stage: our search technology across Rambler&Co products, from Subbotniy and News to search across all of LiveJournal.
Constraint
Many Rambler&Co products needed search, but each team was solving that problem separately
Key decisions
Build search across the whole group's resources on our own technology: a single entry point adapted to each property.
Outcome
Our search in every Rambler&Co product that needs its own search · LiveJournal: about 1 billion documents, more than 5 TB
Context
This was the continuation of our work on Rambler Search. Rambler's main web results came from an external search engine, but the team also had its own search technology, and that is what we kept developing.
Many of the group's products needed search: News, Subbotniy, Finance, Horoscopes, Championat, LiveJournal. Before a shared search technology, products solved this in different ways. LiveJournal, for example, used Yandex search.
The constraint
One technology was not enough on its own, because every property had its own needs. Championat needed team, player and event cards; News needed an easy way back to the news after a search; LiveJournal needed search across a huge archive with blocked pages, reposts and search spam.
And every integration had to be negotiated separately with the product team.
My role
I owned the product and a large part of the internal business development. The team and I went to the group's product teams and offered to take search off their hands so their developers could focus on their own product. I negotiated the integrations, set priorities across properties and worked with each team on how search should look and work for them. Our team built and ran the search.
Key decisions
Build search across the whole group's resources on our own technology: a single entry point adapted to each property.
For the group's properties, an embedded search. It showed the property's own content and, next to it, blocks from other group projects, web search and ads. That gave the group an additional monetisation channel and control over residual traffic.
For products with specific needs, a separate search on the same technology, built around the product: for Championat with team, player and event cards; for LiveJournal a search across the whole service to replace Yandex search.
What changed
In Q2 2016 we launched the embedded search on Subbotniy and built separate searches for Championat, Nightparty and the comments admin tool.
In Q3 the embedded search went live on Rambler/News, with switching between the main and the local search. We built a search for Horoscopes, autocomplete and many improvements for Championat, and started showing LiveJournal results in the embedded search on group properties. By the end of 2016 the embedded search was running on News, Subbotniy, Finance and Rambler ID.
We tuned ranking separately for Championat, Afisha.ru and LiveJournal; as the next stage we planned to do the same for Lenta.ru, Gazeta.ru, Letidor, Motor, Avtorambler, Eda and Finance.
LiveJournal was the biggest technical task. We built search across the whole service, with a whitelist of users in the index, filtering of blocked pages and ranking by post date and relevance, then handed the API to the LiveJournal team to replace Yandex search.
In parallel we moved the searches from Sphinx to ElasticSearch and from FreeBSD to Linux.
Historical product artifacts
Result
In the end our team's search technology was adopted in every Rambler&Co product and service that needed its own search.
In April 2017 LiveJournal launched a public beta of search developed by the Rambler Search team: about 1 billion documents totalling more than 5 TB, on ElasticSearch.
Integrations with the group's properties gave our team access to content from across the group. That later became one of the prerequisites for Rambler&Co recommendations.