uk startup funding digest
i use this as a daily discovery feed — early-stage funding is often the signal to reach out about a role.
the problem
uk startup funding digest is a daily pipeline that finds early-stage uk funding rounds and logs them to a spreadsheet automatically. the problem it solves: tracking that funding by hand means reading dozens of sources every day, and the tools that exist are either too broad — us-heavy, every stage — or sit behind an expensive subscription. there was no lightweight, uk-specific, early-stage-only feed.
the approach
- 01 fetch
every morning the pipeline pulls the last 24 hours of uk startup news.
- a dozen uk-focused news sources are polled in parallel on a daily schedule.
- anything older than 24 hours is filtered out before processing — only fresh rounds get through.
- 02 extract
claude reads each article and pulls the funding round out of the prose.
- for every genuine early-stage round it extracts company name, amount, stage, lead investors and source url.
- articles with no funding news are dropped — the model classifies, it doesn't pad.
- 03 deduplicate
the same round, reported by five outlets, must land as one row.
- a composite key built from company name and amount identifies a round regardless of which outlet reported it.
- a second sighting updates the existing row instead of creating a duplicate.
- 04 log
confirmed rounds are written to a google sheet.
- one clean row per round — company, amount, stage, investors, source — ready to act on.
engineering challenges
- one round, many reports
a single funding round gets written up by half a dozen outlets within a day.
- poll a dozen sources and the same seed round appears again and again.
- an unfiltered feed of duplicates is noise, not intelligence.
rounds are identified by what they are, not where they were found.
- a composite key of company name and funding amount is stable across every outlet.
- the sheet write is an upsert keyed on it — a re-report updates the row, never adds one.
- structured data from unstructured prose
a funding announcement is a paragraph of journalism, not a record.
- amount, stage and lead investors are phrased differently in every article — and most articles aren't about funding at all.
an ai extraction step does the classifying and the structuring in one pass.
- claude reads each article, decides whether it reports an early-stage round, and if so returns the fields as structured data.
- non-funding articles are discarded here, so nothing downstream has to second-guess the input.
the workflow
