SkillGraph-Service: A KG-First Microservice for Competency Framework Integration
Introducing a new microservice, SkillGraph-Service, which integrates the authoritative competency frameworks ESCO, ROME, and O*NET into a Knowledge Graph (KG) that preserves provenance. This system features a KG-first architecture with a fallback to LLM, merging symbolic precision with sub-symbolic adaptability. It employs a hybrid retrieval engine that combines SQLite FTS5 with HNSW vector search to address vocabulary mismatches in educator inquiries, utilizing Large Language Models (LLMs) solely for limited ranking and audience-specific explanations. Empirical tests on a multilingual dataset demonstrate exceptional retrieval performance (nDCG@5>0.94) with latency under 200 ms. The associated paper can be found on arXiv with the identifier 2605.01582.
Key facts
- SkillGraph-Service unifies ESCO, ROME, and O*NET frameworks
- Uses KG-first, LLM-fallback architecture
- Hybrid retrieval engine fuses SQLite FTS5 and HNSW vector search
- LLMs used only for constrained ranking and explanation
- Empirical evaluation shows nDCG@5>0.94 and sub-200 ms latency
- Paper available on arXiv:2605.01582
- Addresses vocabulary mismatch in educator queries
- Designed as an interoperable microservice
Entities
Institutions
- ESCO
- ROME
- O*NET
- arXiv