Advanced data search
We build full-text search, filters and autocomplete over hundreds of thousands of records, with responses in tens of milliseconds. And where the exact word is not enough, search by meaning.
What you get with this service
The nine topics you ask about most. Open whichever interests you – the details and answers are inside.
Czech the way customers write it
Inflection, diacritics, different word order and typos.
We tune the analysers to your content, so “dětské bundy” (kids' jackets) also finds “bunda pro děti” and “detske bundy”.
Search by meaning
We find even a record that does not contain the search word at all.
Vector search works with meaning. It suits documents, listings, knowledge bases or goods that people call by their own names.
- Do we need AI for similarity?
- Vector search uses a model that turns text into vectors, and it can run locally on your side. It makes nothing up – it only searches.
- Does it work for internal company documents too?
- Yes. Someone who does not know the company jargon finds a guideline or a contract by describing the problem in their own words – they do not need to know what the document is called.
Fast filters and counts
Combinations of dozens of parameters, even over a large catalogue.
The counts for each value are recalculated with every selection. We build the index and aggregations so that responses stay in tens of milliseconds.
- Can it handle our volume?
- Hundreds of thousands to millions of records are common. We design the index and the hardware around the data volume and the number of queries at peak.
- What technologies do you use?
- Elasticsearch or OpenSearch for full-text and filters, PostgreSQL with pgvector or a dedicated vector database for similarity search.
Autocomplete
From the first characters, it suggests what people actually search for.
With typo tolerance and only what is available right now, so it never sends anyone to a sold-out product or a withdrawn listing.
Ranking by your rules
Status, freshness, availability and priorities in the order of results.
The rules are documented and you change them without touching the application.
Zero-result queries as data
We show you what people search for and do not find.
It is the most accurate information about what is missing from your catalogue or content. We use it to keep tuning relevance, and you know what to add.
How we build it
Built on your users' real queries, not a lab sample.
-
Data and queries What people search for, how they search for it today and where it fails. Plus the data structure and the language of the content.
-
Index, relevance and similarity Full-text with Czech-language analysers, filters with counts and a vector index for search by meaning.
-
Tuning in production We measure queries with no results and user behaviour, and keep adjusting relevance accordingly.
What it costs
We quote a price after reviewing your data and the queries that find nothing today.
It depends mainly on the volume and structure of the data, the number of filters and whether search by meaning is needed. On a sample of your data we show what the results would look like before you decide.
Operation and tuning
Response time monitoring and ongoing relevance tuning – only if you want them.
The index and relevance rules are documented and run on your side. Proven open source, no black box and no licence to tie you down.
Operation, support and developmentSearch we have built
Listings, catalogues, documents and learning content – tens of thousands to millions of records.
All our references
Škola s nadhledem
Online practice from primary to secondary school – over 6,600 exercises for thousands of pupils and teachers.
- 6,600+
- online exercises
- primary + secondary
- curriculum coverage
MojeBillboardy
Advertising spaces from agencies nationwide on one map. Clients send requests for free and agencies unlock contacts with credit. We built the portal from scratch and have developed it since 2011.
Bezrealitky
The most successful real estate portal for selling and renting property directly from owners. We have been developing it continuously since 2007 and keep it running under the load of tens of thousands of people a day.
Adol Monitor
Monitoring of property auctions and enforcement sales – automatic data collection from dozens of sources, merged into one place, and a watchdog for clients.
-
CK Intertrans
A travel agency website with a booking system connected to the agency's internal system. What began as an add-on to the printed catalogue has become the main sales channel.
- Websites
- Search
-
Grainterminal
A platform for trading agricultural commodities – prices calculated from the exchange and input from traders, counterparty matching on a map and trading all the way to the invoice. Today it is part of CleverFarm, and we continue to manage its trading module.
- Portals
- Internal systems
- Search
- Data processing
- Maps
Why bring it to us
We build search on data we run ourselves. It is tuned on real queries, not on a sample.
- We have had full-text search over listings, catalogues and learning content in production for years – Bezrealitky, Adol Monitor, Škola s nadhledem and Intertrans.
- The result is not decided by the choice of database, but by Czech-language analysers, relevance rules and how the index is kept up to date.
- The index and relevance rules are documented and run on your side, so you are not tied to a licence or a black box.
Let's try search on your data
Send us a sample of your data and a few queries that find nothing today.