Richard Careaga

Analytic Pragmatician

AI is over-hyped, but under-estimated

Seattle, WA

About

Bridging the gap between high-level legal governance and modern technical execution. With a career spent managing regulatory risk at some of the world’s largest financial institutions, Richard provides boards with a rare "bilingual" expertise: the ability to translate complex data architectures and AI ethics into actionable fiduciary strategies. JPMorgan Chase, NA: Associate General Counsel Washington Mutual, Inc: SVP & Associate General Counsel Infrastructure: Expert in large-scale regulatory and environmental permitting. Active researcher and developer in the Julia programming language, specializing in spatial data visualization and algorithmic accountability.

1 Project

AI Assisted Privilege Log Preparation

Web AppSource Release Post

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."

Try It Live

Media Appearances

The Unglamorous Foundation of Data Science

The Diary of a Mad Data Scientist™️

The work of applied statistics and data science cannot begin until the difficult work of data cleaning has finished. In the corporate world, data doesn't arrive on your desk ready to "slice and dice." It arrives trapped in overdecorated spreadsheets, buried in embedded calculations, and riddled with human error. In this episode of The Diary of a Mad Data Scientist, author and retired lawyer Richard Careaga joins host Anne Turner to discuss the "least glamorous" but most critical role in the field: The Data Janitor. Using a real-world case study from "MegaCo"—a multinational facing a 7-billion-euro spend visibility problem—Richard illustrates how algorithmic "munging" and "scrubbing" can turn a mountain of digital junk into a high-level tool for cost-targeting and negotiation. In this video, you’ll learn: Why spreadsheets are the "razzle-dazzle" traps of the finance world. The 8 types of human error that compromise your datasets. How to use Pareto plots to find the 20% of suppliers that account for 80% of your spend. Why data cleaning isn't "scut work"—it’s the foundation of truth.

Philosophy

"It's the second derivative that kills you"

It's not easy but we can adapt to change. What's hard to adapt to is the rate of change.