Case Studies / IT demand monitor

Internal tool · ETL

Public signals in 20+ languages,
one readable dashboard

We built this monitor to find organisations that publicly need IT work: tenders, digital transformation news, new CTOs, IT hiring. Sources write in their own language, and the team reads every card in English or Russian.

ETL pipeline

From a tender notice to a scored, translated card

The monitor runs continuously on local hardware. Every stage keeps its own queue, so a slow source or a busy model only delays its own work.

Extract

Keyless public sources: EU TED tenders for IT services, UK Contracts Finder, CanadaBuys open data, news feeds in 21 languages, Wikidata and a polite crawler for company sites. A site that blocks the crawler is marked and left alone.

Normalise and deduplicate

Organisations are matched by name and domain, so one company found in a tender and in the news becomes one record. Budgets are converted to euros with ECB reference rates. E-mails and phone numbers are removed from stored text.

Classify and score

Rules score the evidence by signal type and freshness, then a local model judges organisation type and IT need. A record is kept only if the organisation's name appears in the evidence, and every rejection has a recorded reason.

Translate

A worker detects the source language and sends names, reasons and evidence titles to our local translation service in batches. Originals are never overwritten, and each translation is checked before it is saved.

Load and serve

Accepted organisations go into the catalogue and a live dashboard with filters, tender details and project values in euros. New items reach the screen over WebSocket without a reload.

Original one click away

Each card shows the translation with the source language marked, and a toggle switches the whole card back to the original. A text still in the queue is shown as original with a note, never as a guess.

Verified and planned

What runs today and what is still a plan

Running

Continuous collection from tender and news sources, deduplication, local model scoring, translation into English and Russian, tender detail pages and the dashboard. It is an internal tool, not a public service.

Not yet

Some sources need a dedicated adapter before their full detail can be read. Short or mixed-language texts are not translated when the language cannot be detected reliably.

FAQ

About this case

No. It is our internal tool for finding companies that need IT work. No client data or results are shown here.
Local models only, for scoring and for translation. No cloud LLM or paid search API is called.
E-mails and phone numbers are removed from stored page text and no person names are extracted. Records keep role hints such as Head of IT.
Yes. Sources and query templates are configurable per country and language, and each new source is an adapter with its own log.

Need a monitor for your market?

Tell us which signals matter and in which languages your sources write. We reply within one business day.