DataGardener · UK company intelligence · 17M+ companies
Search 17 million companies by asking a question.
DataGardener holds financials, funding, headcount, directors and risk signals for every UK company. Finding the right slice meant hunting through dozens of filters, and sometimes emailing support to ask where a filter lived.
We built a retrieval layer that turns a sentence like “turnover above £10M and more than 500 employees” into the exact JSON query their backend expects. A vector store of worked examples guides the model, so it writes queries by analogy rather than guessing.
2–5 minper search before, by hand through filters
30–60 minwhen users had to ask support where a filter was
Secondsnow: one typed question, one table
The AI search runs inside DataGardener's paid product, so it's available to their subscribers.
ClaudeLangChainVector storePythonJSON query schema
How a question becomes a query
1 · Question“Companies in Manchester with turnover over £10M and 500+ staff”
2 · Nearest worked examples
“firms with more than 1,000 employees”employees.gt
“turnover between £5M and £20M”turnover.range
“companies based in Leeds”region.eq
3 · Query written by the model, checked against the schema{ "region": "Manchester",
"turnover_gbp": { "gt": 10000000 },
"employees": { "gt": 500 } }
"turnover_gbp": { "gt": 10000000 },
"employees": { "gt": 500 } }
4 · BackendRuns the query and returns a sortable table, exportable like any other search.
