Have you tried using non-LLM based methods? Like starting with something rules-based and working through a layered multi-model setup?
That’s what we’ve been using for document extraction where accuracy needs precision (capital markets documents, medical assessments). We had a go at pure LLM with medical documents but the output was poor and felt like it would take substantial investment to create something more robust.
That’s what we’ve been using for document extraction where accuracy needs precision (capital markets documents, medical assessments). We had a go at pure LLM with medical documents but the output was poor and felt like it would take substantial investment to create something more robust.