CareProfSys

A smart career profiler: take a person’s education, interests and public traces, map them onto a labour-market ontology, and recommend occupations — and the learning path that would make those occupations reachable.

Career counselling COR / ESCO Published research

Desk with notebooks of occupation trees and a career-mapping interface on a laptop

Context

Career advice still leans on static lists and counsellor memory. Occupations in Romania are already described — in COR, and in Europe through ESCO — but those nomenclatures are not a product a student can talk to. Industry 4.0 also keeps minting roles that a printed framework is slow to absorb.

What we built

A semantic fusion system: profiles from education, psychological tests and public data; a five-level occupation ontology aligned with COR and ESCO; a recommender that scores a person against a profession and can also invert the question — which courses close the gap.

Outcome

The concept was implemented and observed in a university career-counselling environment. The ontology, the engineering-focused career model, and the recommender are published. A related strand looks at chatbot-based training as a way to keep professional education going once a path is chosen.

The ontology

Occupations sit in a five-level tree. The first four levels follow the same digits as COR and correspond to ESCO / ISCO groups. The fifth level holds roles that exist on the Romanian market and do not have a European code. Each occupation carries the COR sheet: general activities, context of work, working style, values, skills, knowledge, aptitudes and interests. Study domains are a second pillar, so education and work are in the same graph rather than two spreadsheets.

Queries over that graph — SPARQL against a graph store, exposed over REST — can answer “given these qualifications, which professions are open?” and “given this profession, what knowledge is still missing?”. Similarity between a profile and a role is computed on the ontology path, not only on keyword overlap.

How a profile is built

The system does not wait for a perfect CV. It fuses what is already there: formal education, tests, and traces from public professional networks where they exist. A Europass CV can seed the same profile. Machine learning — including transformer models for language — sits next to the ontology so unstructured text about skills can be aligned with structured occupation requirements.

The stack around that idea is the one a counselling product actually needs: Java and Spring on the services, React on the interface, GraphDB and SPARQL for the knowledge base, MySQL for operational data, Python, TensorFlow and Django where the models live.

Recommendations, both ways

One direction is a job: occupations that fit the person as they are. The other is a learning path: if someone wants a specific role, the system can point at the knowledge and education that still stand between them and that role. That second direction is what makes it counselling rather than a vacancy search.

A related publication looks at chatbot-based training for project management — whether conversational systems are a side channel for corporate training or a durable way to keep professional education available. It sits in the same research line: once a career is recommended, the person still has to learn.

Visit careprofsys.upb.ro