Where the work gets stuck
It usually starts with one of these:
- Analysts repeatedly gather the same filings, transcripts, notes, and portfolio data.
- Internal research exists but is hard to find or reuse.
- A recurring review (a new deal, quarterly earnings, or a portfolio screen) still depends on manual steps that vary by person.
- People get good results single-player, but none of it is shared or connected to firm data.
We pick one workflow together and I trace it from the source material to the finished output. Then we tease apart what the models can do, what plain code should do, and what stays with the analyst.
I build the smallest version your team can test. Sometimes the answer is a process change or plain code, and I'll say so.
Good first workflows are an hour of gathering and fifteen minutes of putting it together: reviewing earnings transcripts, prepping for a management meeting, refreshing a comp table. Some checks only become realistic with these tools, like comparing this year's 10-K to last year's, footnote by footnote.
What I've built
I'm building screening, report generation, and research search for an investment team, and wiring their data into the tools they already work in.
Before PragmaNexus, I spent most of my career at investment firms and fintech companies. I built derivatives trading software for a hedge fund, the platform CircleUp used to score companies for investment, and the full stack for a quantitative macro strategy.
Five Lessons from Putting AI Into Research Teams
From a talk to a private working group of finance professionals: which tasks to automate, where analysts push back, and how to tell if a workflow is getting worse. Interactive version
Invest with AI: Deploying AI on the Buyside
I joined Brett Caughran and Khe Hy on Invest with AI to talk about moving from individual experiments to shared workflows. Most of it was about process and communication, which in my experience is where firms get stuck more than the technology.
How engagements work
Most build engagements are monthly retainers. A scoped assessment is also possible.
If you have engineers, I work alongside them. If you don't, I cover the technical side.
It works best when there's one workflow to start with and the analysts who run it can try each version as we go.
Start with one workflow
Tell me how the workflow runs today and where it slows down.