Second practice area · Part 3 of 3

Turning Coffee Data into Coffee Knowledge

An invitation to build a public Coffee Knowledge Vault on ontology, knowledge graphs and AI — see a simulated dashboard of the vault in use

Neo Gens · August 202613 min readShort version for growersSee the demo dashboard

08How a commons stays a commons

Knowledge commons fail in predictable ways. They get captured by one large contributor, or they fragment into incompatible forks, or they fill with unverifiable junk, or the stewarding organisation quietly turns them into a product. Any serious proposal has to say how it intends not to do that.

Our commitments are four.

The structure is open.

The ontology, the schema and the reference vocabularies are published. Anyone can implement against them, extend them, or argue with them. A commons whose structure is proprietary is not a commons.

Contributors keep rights over their own records.

A farmer's farm record is theirs: portable, exportable, shareable with the buyers they choose, revocable from the ones they do not. This is where the ownership question belongs — as a governance rule that keeps participation voluntary and keeps the vault honest, rather than as the headline argument.

Provenance is mandatory, everywhere.

Every fact in the vault carries its source and its verification status. That is what prevents both capture and junk: a large contributor cannot overwrite the record, they can only add claims that stand or fall on their evidence.

Stewardship is separated from ownership.

Neo Gens funds and maintains the work. Neo Gens does not own the knowledge, and the design is deliberately unattractive to acquire, because the valuable part is published.

None of this is charity, and we should be plain about it. We think a functioning knowledge commons makes the whole specialty coffee sector more productive, and organisations that know how to build and use structured knowledge will do well in that world. That is the business we are in. But the vault itself has to be a public good, because a vault that one company owns is just a database with better marketing, and nobody should deposit into it.

09Proof: the first deposits

Global ambitions deserve local evidence, so here is an honest account of where the work stands.

We chose Thailand for the first deposits for a reason that surprised us. Thailand grows roughly 16,000 tonnes of coffee a year and drinks around 95,500 — it imported about 80,000 tonnes in 2025 and exported some $9.2 million worth of green and roasted coffee in 2024, ranking 87th in the world. Thai smallholders sell domestically, into a market that wants six times what the country produces. EUDR compliance is not their problem.

That made Thailand a better test, not a worse one. If a knowledge vault only attracts contributions when a regulation forces them, it is a compliance product. In Thailand the only reason anyone would contribute is that the vault gives back more than it takes. That is a much harder bar and a much more honest one.

What exists today: a formal domain ontology built on published academic work on Thai coffee knowledge and extended with supply-chain and data-integrity modules; a Neo4j schema with Lot as the central linking entity; two flavour vocabularies — the international wheel and a Thai/Asian wheel — connected by curated mappings; an eight-stage capture pipeline design from a messaging app to a validated graph write; the claim-and-evidence provenance layer described above; and a competency-question register in which six priority questions are implemented as queries and pass their evaluation tests against a demonstration dataset. The first tangible output, a one-page lot profile for a roaster, has been generated end to end. The ontology is built federation-ready, so a farm registered in Colombia, Ethiopia or Honduras joins the same structure as one in Chiang Rai.

What does not exist yet: contributors. The pilot cooperative, the field interviews that will test our competency questions against what growers actually ask, and the cupping sessions that will validate the Asian flavour wheel are the next phase, not a completed one. Every question in the register is still marked draft, precisely because it came from research rather than from a farmer's mouth. We are not promoting any of them until we have sat with people who grow coffee.

We would rather publish that distinction clearly than round it up. The sector has enough pilots described in the past tense.

10An invitation

A vault with one contributor is a database. What we are asking for is the first cohort.

  • Cooperatives and producer organisations willing to make the first deposits in a second origin, and to tell us what the structure gets wrong about how coffee actually works where they are
  • Roasters who will contribute cupping data and use farmer-sourced profiles in real sourcing decisions — and say plainly when the profiles are not good enough to use
  • Mills and exporters who see the case for depositing processing and compliance data once instead of documenting the same farms every season
  • Researchers, institutes and origin bodies with agronomic knowledge or a regional flavour vocabulary that belongs in a shared structure rather than in a PDF
  • Development organisations and funders interested in knowledge infrastructure as a complement to the harder work on prices and contracts, not a substitute for it
  • Ontology and knowledge-graph practitioners willing to read the schema and argue with it

The coffee industry already pays for all of this knowledge. It pays several times over, in duplicated surveys, unprovable quality claims, sourcing decisions made on thin information, and expertise that retires with the person who held it. The only thing missing is a shared place to put it and a structure that makes it compound.

That is a thing a group of people can decide to build. We would like to help build it, and we would like company.

Coffee KM is a public-goods project stewarded by Neo Gens, a Modern Knowledge Management practice. The ontology, schema and vocabularies are published openly. To contribute, get in touch.

Sources

Take part

The pilot needs its first deposits.

If you would like to put the first records into the vault, leave your details and we will get in touch.

Received. We reply within two business days.

Goes to hello@neogens.co — or write there directly.

Get in touch
Request a briefinghello@neogens.co
Previous
Why this is possible now and was not before