What I've built

Investment research platform

1. The problem

   An investment team needed a more consistent way to gather and reuse research spread across several systems.

2. The work

   I sat with the analysts to learn how they work, then built screening, reporting, and search tools around that, wired into the software they already use.

3. What changed

   Analysts can run screens, generate reports, and search firm research without switching tools.

Related: AI in Research Teams

Coding agents across an engineering team

1. The problem

   Each engineer had their own way of giving coding agents context about the codebase, and none of it was shared.

2. The work

   I met with each engineer to see how they worked with agents, then built shared context for the whole team: instructions, skills, and architecture docs that update every week. I also wrote code alongside the team.

3. What changed

   The shared instructions and docs live in one place every engineer's agent can read, and the docs stay current on their own.

Production ML pipeline assessment

1. The problem

   A lending company wanted an outside review of how its team changes and ships production ML models.

2. The work

   I interviewed the data science and engineering teams, traced a representative change through the system, and documented the current workflow and its bottlenecks.

3. What changed

   The team got a written plan that lists what to change, why, and what to do first.

Document generation system

1. The problem

   A healthcare technology company needed to turn long source documents into drafts that experts could review.

2. The work

   I used AI to pull content from the source documents and ordinary software to assemble the output the same way each time. Then I built tests that compare drafts to examples written by experts.

3. What changed

   The team has a repeatable document workflow and a way to measure output quality.

Predictive scheduling model

1. The problem

   A logistics technology company wanted scheduling predictions running live inside their product.

2. The work

   I trained a set of models on historical data, tested them against past outcomes, and deployed them as a live service the product calls.

3. What changed

   The product requests predictions in real time. Before, they came from offline analysis.

Applied AI workshop

1. The problem

   A healthcare organization wanted AI training that fit what the team does.

2. The work

   I built and ran a hands-on session, with exercises from their day-to-day work.

3. What changed

   The team left with the exercises and follow-up materials to reuse.