Neuroaktywni Foundation

The challenge

The Neuroaktywni Foundation supports neurodivergent people. Like most non-profits, it runs on a limited budget and a small team, and its work depends on grants and subsidies.

The problem is structural: calls are scattered across dozens of portals (EU funds, government programmes, municipal competitions, private foundations), announced irregularly and often with short deadlines. Tracking them manually is a full-time job nobody at the foundation has.

The outcome is always the same, and it applies to almost every small organisation: missed calls nobody heard about in time, and applications written in a rush during the last week before a deadline. Every hour the team spends combing through portals is an hour taken away from the people they support - which is what the foundation exists for.

What we built

An AI agent tailored to the foundation’s profile, working in four steps:

1. Source monitoring - the agent scans grant and funding portals on a schedule, instead of waiting for someone on the team to find time to go around them
2. Profile filtering - it discards calls the foundation doesn’t qualify for (legal form, area of activity, required own contribution) and keeps the ones genuinely matching its statutory goals. That’s where the real saving is: not finding calls, but screening out the ones that would have been rejected anyway
3. Deadline alerts - notifications arrive with enough lead time to actually prepare an application, not on the day the call closes
4. Draft documents - for every matching call the agent prepares an application draft from the foundation’s own data: its activity description, past projects and beneficiary groups

A human stays in the loop

The agent does not submit applications. It prepares a draft that a human reviews, completes and approves.

That was a deliberate design decision, not a technical limitation. A grant application is a document the foundation is accountable for in front of the funder - and a mistake in it isn’t a typo, it’s a rejected application and a burned call you wait a year to see again. The agent shortens the path from blank page to finished draft; the decision to send stays with the team.

Why the agent runs on Cloudflare Workers

“24/7 monitoring” sounds like a server running non-stop - and a server running non-stop bills every month whether or not it is doing anything. For a foundation that isn’t a technical detail, it’s a budget line competing with the work the organisation exists to do.

This agent doesn’t run continuously. It wakes on a schedule, walks the sources, stores the result and stops. Cloudflare Workers bills that pattern per invocation rather than per hour of standing still - between one run and the next, the infrastructure costs close to nothing. The data the agent builds drafts from - the activity description, past projects, beneficiary groups - sits in PostgreSQL, so nobody re-enters any of it for the next application.

This wasn’t optimisation for its own sake. In an organisation funded by grants, the cost of running a tool is part of the question of whether the tool makes sense at all - exactly like the payback maths we run for a commercial client.

Results

The foundation no longer misses calls, and every application starts from a ready draft instead of a blank page. The team can spend the recovered time with the people they support.

What this means for you

If someone in your organisation regularly combs the same sources looking for opportunities - grants, tenders, listings - that task is a ready candidate for automation. The biggest saving is usually not the search itself, but filtering out what would have been rejected anyway, and starting every process from a draft rather than from zero.