KOIOS Master Data
Knowledge Graph Engineer - Winchester
Get notified of more jobs like this · No spam, ever
How your CV stacks up
Upload your CV to see how well it fits this job role
?%
About KOIOS
KOIOS is a Winchester company building a universal semantic layer for industrial product data. Our concept dictionary holds over 400,000 concepts drawn from international standards, and we hold IRDI Authority status — we are entitled to issue the globally unique identifiers that make machine-readable master data possible. Our founder, Peter Eales, is the project leader of both ISO 8000-110:2021, and ISO 29002:2026.
KOIOS was founded in 2017, and the platform is in commercial deployment with enterprise clients today. We are contracted to a global industrial distributor to rationalise their product data model and onboard supplier catalogues across thousands of manufacturers — a programme already running across multiple workstreams, against a catalogue taking on hundreds of new product classes every month.
We are building the team accordingly. This is one of several roles opening across the dictionary, the graph and the platform, as we increase the pace at which we deliver. The distributor programme is the forcing function for this one.
The problem you would work on
A distributor lists a product as "Terminal Covers." A European classification standard lists "Cover cap for rail terminal bar." Are those the same concept, or is one broader than the other, or are they merely associated? Multiply that by hundreds of thousands of classes across dozens of vocabularies in fourteen languages, and you have the problem.
Naive string matching fails immediately. So does naive embedding similarity — "fire partitioning" and "firewall" sit close together in vector space and refer to a fire-resistant building barrier and a network security device respectively.
Our current approach treats each concept as a distribution over its terms and definitions and computes an optimal transport cost between distributions, rather than taking a single nearest neighbour. That ranks candidates. A language model then adjudicates the relationship — equivalent, broader, narrower, partially overlapping, associated, unrelated — with a confidence score. A human reviews the output in a dashboard and decides what enters the dictionary.
Reasons to use Rodeo
Start with a chat, not a search bar
Grad scheme, placement, apprenticeship? Not sure what you want yet — that's fine. Your agent talks it through with you and turns "I have no idea" into a shortlist.
It searches the market for you
Every day your agent scans the market matching roles against what actually matters to you, not just keywords on a CV.
Only hits
No noise. No "maybe this fits." Just roles with a clear explanation of why they're right — and where to focus when applying.
What the work looks like
- Designing and improving the matching method — the transport formulation, the embedding layer, the ranking, and how the model adjudication combines with it.
- Building the evaluation harness: ground-truth sets, precision and recall by product class and relation type, calibration, drift detection, regression gates before anything ships.
- Modelling the graph. Concepts, relationships and provenance live in Neo4j alongside MySQL; you would shape the schema and write the queries that the dashboard and the platform run against.
- Turning a working pipeline into something that runs unattended across millions of records.
- Working closely with the CTO, who approves what enters the dictionary. You produce the evidence; he makes the call.
What we are looking for
Essential
- A strong degree in mathematics, statistics, physics, engineering or a comparably quantitative subject. We care more about mathematical maturity than about the label on the degree.
- Real comfort with the mathematics underneath: probability, linear algebra, optimisation, and the ideas behind distance and similarity in high-dimensional spaces. You do not need to have met optimal transport; you do need to be the sort of person who can read about it and then have an opinion.
- Programming ability sufficient to run your own experiments end to end. Python is what we use.
- Intellectual rigour, by which we mean specifically: defining the measurement before running the experiment, and reporting what the numbers show rather than what you hoped they would.
- The ability to work on an underspecified problem. Nobody will hand you the evaluation criteria; deciding what "right" means here is part of the job.
- The right to work in the UK, and the ability to work from our Winchester office.
Get help with your application
Your very own career expert that helps elevate your application to the next level.
An advantage, but genuinely not required
- Neo4j, Cypher, or any graph database experience.
- Embeddings, vector databases, retrieval, or hands-on work with LLM APIs.
- MySQL and relational modelling.
- Optimal transport, Wasserstein distances, or any exposure to metric geometry.
- Any background in taxonomy, ontology, terminology or linguistics.
What we offer
- Ownership of a genuinely unsolved problem with immediate commercial consequence, in a company at the stage where that ownership is real rather than nominal.
- A research thread — we work with the University of Southampton on the underlying graph theory, and there is scope to publish or to take this into postgraduate study.
- Direct work with the author of the international standards this field runs on.
- Equity through our EMI share option scheme, alongside salary.
Location and working pattern
Based at The Pump House, Garnier Road, Winchester. Applicants should be within commuting distance.
How to apply
Send a CV and a short note to jonathan.grant@koiosmasterdata.com.
We are happy to hear from people who do not meet every line above.
“It took my CV and asked me questions relevant to understanding what kind of jobs to suggest for me. Suggestions were almost perfect. Jobs were exactly what I’ve been looking for.”
Jessica, London
Skills