A 150M model that extracts verbatim evidence spans for RAG, no LLM call

A 150M-parameter model extracts exact, verbatim evidence spans from documents for retrieval-augmented generation (RAG), eliminating the need for costly LLM calls. It is designed for developers and researchers building efficient, fact-grounded AI systems where precision and low latency are critical. This is interesting because it achieves high-accuracy evidence retrieval with a fraction of the compute, enabling scalable, citation-backed answers without reliance on large language models.