AI Assisted Privilege Log Preparationflagship
The Enron corpus of 500,000 plaintext emails from 142 document custodians is a natural candidate for LLM assisted litigation preparation. Recent changes to the U.S. Federal Rules of Civil Procedure accelerate the meet and confer to a much earlier point. Over the course of an afternoon, Claude Opus 4.6 was guided through a preprocessed, deduplicated version of the corpus. Careaga, R. (2026). Preprocessed Enron corpus [Data set]. Zenodo. https://doi.org/10.5281/zenodo.18857652 The LLM extended. my previous network/semantic analysis to identify and classify communities of email users and identify tiers most likely to contain attorney-client communication or work product privileged records and the large residuum where sampling could serve. I am currently extending that work with DiscoveryGraph.jl to implement human review where attorney judgment is required. I'm finding that the algorithms so far appear overinclusive, bringing in administrivia such as "not able to attend weekly litigation update meeting."