Semantic Band-Gap (LLM + KG)

NL→KG with normalization; metada for provenance.
NL→KG with normalization; metada for provenance.
RDF/SPARQL LLM-assisted retrieval/query llama3.2:3b local

What: Built a semantic knowledge graph (KG) for custom-featurized semiconductor band-gap data using RDF triples.

Why: Enable structured, explainable retrieval and easy integration of heterogeneous materials datasets.

How: Designed an RDF schema (materials, composition, structure-features, measurements); ingested sources, provenence metadata to RDF; used a light-weight local Ollama model to translate natural language: into executable SPARQL queries (secondary purpose) or into deduped + sanitized ingestible triples to the KG , supporting url, pdf, raw/fabricated texts [PRIMARY TASK ].

Results: A queryable single source of truth that improves accessibility, automation, and reproducibility for band-gap analysis.

GitHub

Ontology
Ontology used in the mini MSE KG—Material with five literal properties.
Pipeline
Guardrailed NL→SPARQL + typed row normalization