Need & Know - SCC, a Swiss nonprofit in formation

Context

Sustainability platforms record what organisations are doing. The UN SDG Actions Platform holds roughly 8,800 entries, and none of them has a field for what an organisation needs. Need & Know starts from the other side. An NGO describes the problem it is stuck on in its own words, and the platform suggests research and expertise that could help, each with a plain-language reason. It is planned as a Swiss nonprofit, starting with climate adaptation in Suisse romande. Beyond a clickable prototype with simulated AI, nothing exists yet.

The real problem

Matching was never the hard part. The research side can be drafted from open data before anyone signs up: publications, funded projects, researcher identities. That is where the difficult problems sit. The same work arrives as a grant, a project and two papers and has to become one record. A researcher who corrects their record must never be overwritten by next week's import. And pre-populating someone's work without asking only stays honest if every field can say where it came from.

My role

Co-founder for product and technology, alongside a co-founder from the sustainability field who leads partnerships. I framed the problem and the service model, then designed the back end: system context, containers and stack, data model, the four key flows, the API surface, non-functional requirements and a staged cost estimate, drawn up in 19 architecture diagrams and a UI concept. Claude was my drafting and review partner throughout. No developer is hired, and none of the back end is built.

System-level decisions

Fourteen decisions are written up as architecture decision records, each with the alternatives set aside and what it costs if it turns out wrong. Three carry the design.

One Postgres instance holds the relational data, full-text search, geography and vectors. At pilot scale a graph database would buy traversal speed the product won't need before 2028, at the price of a second system to run.

Provenance is tracked per field. An import may overwrite a value that came from a source. It may never overwrite one that came from a person. That one rule is what lets the platform draft a researcher's record without impersonating them.

No match score is computed for display. Factors are stored and explained, because a percentage invites trust without reading the reasoning. The early prototype showed one; the architecture now rules out any endpoint that returns it.

Outcome

The architecture pack asks its reviewers to attack it. The first review, run with Claude Code, concluded it should not be built as specified: two languages are a heavy tax on a one-developer nonprofit, the per-field provenance model is untested at volume, and the pilot estimate of 35 to 50 developer-days is optimistic. Those objections are open. They are the right ones to have before a grant application is sized on that number.

The larger limit sits upstream of all of it. The plan is ten conversations with NGOs before any code, to learn whether they can describe a need that is matchable at all. As of mid-September none had taken place, and a forestry specialist we consulted doubted the need exists. Until those conversations say otherwise, this stays a concept.

Website: www.needandknow.com