28 relevant opportunities for a think tank — out of 10,000+ pages processed

GGTC Kyiv

Client: GGTC Kyiv — the Global Government Technology Centre, the world's second GovTech centre after Berlin
Website: kyivgovtechcentre.org
Industry: GovTech · digital transformation of government
Who uses it: The centre's fundraising lead
Format: Grant-scouting agent + a web interface for the team
Timeline: First working result — under a week from kickoff

The problem

One person was hunting for grants. More than 35 sources: grant platforms, embassy pages, sector newsletters, Telegram channels — and the whole round had to be repeated every week, because new calls appear all the time. Every call is 40+ pages of terms: budget, deadline, eligibility, consortium requirements. You have to read all of it to answer one question — does the organisation even qualify? Most of the time it doesn't.

What it does

We built the centre Granty — an agent that does the round itself. On a schedule it reads every source, scores every opportunity against the centre's own criteria (focus, budget, deadline, eligibility, consortium) and surfaces a short list with a 1–10 score and a reason each grant made it. Sources and criteria live in the centre's own Google files: the fundraiser changes the rules herself, and the agent shows the diff before applying them. No developer needed for that.

One screen instead of dozens of tabs: candidates, recommended ones (≥7), a manual run button and the interval field — the centre decides how often the agent goes out. Below it, the last run: how many sources were covered and what came back.

Covers 35+ sources: grant platforms, embassy pages, Telegram channels — and takes European calls from the official EU Funding & Tenders API rather than scraping
Scores every opportunity against the centre's seven criteria and shows the breakdown — you can see why it got that score instead of «the AI said so»
Criteria live in the client's own Google Doc: edit → the agent shows a diff → confirm. The previous version keeps running until the new one is approved
Results land both in the interface and in a Google Sheet; «reviewed / rejected» triage happens right in the list
Sources that failed to load don't vanish — the agent lists them separately for a human to check by hand

The score breakdown: how many points each requirement gave — mission, budget, eligibility, geography, donor, deadline, consortium. The call description and the «why it fits» reasoning are redacted: that is the centre's grant pipeline, and it stays with the centre. Next to it, the note field the team uses to calibrate the agent.

relevant out of 420

One run: 35 sources covered, 420 opportunities read, 28 surfaced with an explanation.

Minutes

for a full round of sources

The same reading by hand — roughly 160 hours by our estimate, about a month.

week to first result

From kickoff to a working agent with live findings — four days.

What the fundraiser is left with: a short list with the score, budget, deadline and source. «Reviewed» and «Reject» sit right there. The call titles are redacted — these are the applications the centre is working on right now.

28 relevant opportunities for a think tank — out of 10,000+ pages processed

Related solutions

Grants, but a human isn't the one searching

Tenders and opportunities, but somebody is monitoring them

Searching past matters, but in seconds instead of hours