rare impact fund toolkit

Non-profit organisations are being asked to adopt AI faster than the sector has had time to learn what safe AI adoption looks like in their context. The Rare Impact Fund toolkit is a small commission to build the instruments that close the gap, tools the partner organisations can run themselves, with the careful defaults already in place.

The Rare Impact Fund supports a portfolio of mental-health non-profits whose work spans clinical adjacent care, peer support, and capability building. Each is fielding the same set of AI-adoption questions and producing the same set of partial answers. The toolkit is an attempt to give the portfolio a shared substrate.

Six instruments are in scope. An impact-assessment visualiser, an evidence-freshness auditor, training-material adapters for different audiences and reading levels, an outcomes-tracking dashboard, a safety-guardrails harness, and a small evaluation suite. Each tool is built once, in the open, and then deployed to the portfolio organisations with their own data and their own boundaries.

The architectural assumption is that non-profits do not want (and should not need) to run their own ML team. The tools ship as web applications with the model calls handled server-side and the safety-relevant logic kept legible to the organisation that owns it.

Early in the build, but the most useful artefact so far has been the evidence-freshness auditor: a small tool that walks a non-profit's public-facing claims, finds the citations behind them, and reports back which sources have moved or been superseded. Several partner organisations have used it as the trigger to refresh material that had been quietly drifting for years.

In active build. Two tools are in pilot inside Rare Impact Fund partners; the remaining four are sequenced for delivery over the next two quarters.