A record of AI systems I have actually built and put to work: inside a legal department, across a company, and in graduate coursework at Columbia. The through-line is the one Blackletter runs on. The judgment stays with the lawyer, and the tools carry it.
As general counsel and chief compliance officer of a heavy-civil construction firm, I have not advised on AI adoption from the outside. I have run it.
Led the AI transformation of the in-house legal department. Built contract-review "skills" in the Claude Cowork environment that apply the department's own markup standards to incoming documents, so the work that comes back looks the way the department produces it rather than the way a generic model would.
Drafted the policies and procedures governing how the company's roughly 800 employees may use AI. Founded and led a Claude Cowork pilot program, and built a range of agents and workflows around the tasks people actually do across the business.
Selected and implemented enterprise HRIS software for the organization, then led its rollout and ongoing governance. That work ran from the vendor evaluation through implementation to the rules for how the system is maintained and used.
Currently leading a digital transformation for a real estate management company, bringing the same playbook-first approach to a new organization and a different set of workflows.
An executive MBA at Columbia Business School, on the Dean's List, with coursework concentrated in applied AI and machine learning. Selected work:
Designed a Retrieval-Augmented Generation deployment plan for a heavy-civil contractor: a centralized, searchable policy repository that lets field managers ask policy questions in plain language, with the compliance department as the clearinghouse for what goes into it. The plan covered the business case, stakeholders, rollout, roadblocks, and governance.
Built a machine-learning system to match job candidates to roles for a hiring-technology startup. The work ran on text embeddings, feature engineering, and model evaluation to score how well a candidate fit a given job.
Applied machine-learning modeling across regression, classification, and neural networks, with hands-on feature engineering and model comparison.
Artificial intelligence and immersive technology, and where they fit into how a business operates.
Programming foundations for working with data in Python.
The real engine isn't code or an LLM. It's the lawyer's knowledge driving it.
From "human (lawyer) in the loop"
I write "human (lawyer) in the loop," a Substack on how lawyers can put AI to use without handing over their judgment. Several posts document small tools I built to test the ideas rather than just assert them.
A customized indemnification-clause editor built on the OpenAI API for less than a cent per run. It reads a counterparty's markup, matches the edit to a scenario in my playbook spreadsheet, and returns both marching orders and a revised clause.
Read the buildA head-to-head test of the spreadsheet-backed editor against a general-purpose model, to show where a documented playbook beats raw fluency. The DIY tool held its own by retrieving the human-crafted response once it matched the scenario.
Read the testA proprietary case-annotation tool built in the Claude Cowork environment for a live matter, and a candid write-up of where it helped and where it needed a lawyer watching over it.
Read the reviewThe idea underneath all of it: taking the "art" of practice, the market-specific negotiation moves a model does not know, and reducing it to a structure a tool can actually use.
Read the pieceIf any of this maps to what your practice is trying to do, I would want to understand how you work before recommending anything.
[email protected]