on the record
Published analysis and independent writing on quantitative finance, financial data engineering, and the practice of building systems that reason about markets.
how databricks can unlock advanced analytics for every business user
The gap between data engineering teams and business analysts has been a structural problem in enterprise analytics for a decade. Databricks' Lakehouse architecture — and its surrounding tooling ecosystem — offers a credible answer: a single platform where raw data ingestion, transformation, ML experimentation, and business-facing reporting coexist without friction. This piece examines how the platform's design choices lower the floor without raising the ceiling, and what that means for teams that have historically been locked out of advanced analytics by engineering dependencies.
Read the Article →
- Publisher Rearc
- Year 2025
- Category Data Analytics
- Platform Databricks
your best broker is probably your worst broker. depends on the order.
VWAP is the most-cited benchmark in execution analytics, and the most misunderstood. Beating VWAP on a single trade says almost nothing about whether your execution was good — it says something about whether the market moved with or against your direction. This piece examines what slippage relative to VWAP can and cannot reveal, how to read it alongside arrival cost and participation rate, and why the framing matters when evaluating broker performance at scale.
Read more →
i built this site with two ai tools. one got cut.
Before writing a single prompt, the process started with an audit — a full design assessment of what a good site actually needed. That spec went first into a fast-generation tool, then got rebuilt properly once its limits showed. Notes on why front-loading the thinking mattered more than any individual prompt.
Read more →
breaking an ai job search system… and making it even better
Six weeks of scoring job postings I never applied to — and the log of passes turned out to be the most useful output. Notes from QA-testing an AI job search kit against my own search: guardrail tests, calibration drift at the edges of a rubric, and why a do-not-overstate list written in advance beats judgment made under pressure.
Read more →More on the way
want to discuss any of these ideas in depth?