SLED

SLED—Semantic LLM Enrichment of Data—uses language models to identify and document the meaning of scientific data fields. The system is designed to reduce one of data science’s most persistent bottlenecks: understanding what unfamiliar columns actually represent.

From labels to meaning

Real-world datasets rarely arrive with complete documentation. Names such as id, value, or location are ambiguous without context. SLED combines column names, values, surrounding metadata, and language-model reasoning to infer semantic types and make those inferences inspectable.

Why it matters

Semantic typing supports validation, interoperability, documentation, and reuse. By automating part of that work, SLED helps researchers move toward FAIR data practices without adding another manual annotation burden.

The project began in Andy Yu’s master’s research and continues SAIL’s broader work in language models for document and data intelligence.