
How an Agent Keeps My Research Organized
Meeting notes, a decision log, an operating manual for each experiment, and a checked package for supervisors. The agent does not do the research. It keeps the research findable.
Older posts on data science and data engineering, revised for today.
Between 2017 and 2019 I kept a technical blog while I worked as a data scientist. These are the pieces from it that still say something useful, rewritten in 2026 with the dated parts explained and a note at the end of each on what I would do differently now.

Meeting notes, a decision log, an operating manual for each experiment, and a checked package for supervisors. The agent does not do the research. It keeps the research findable.

How I keep three annotated bibliographies current with a search API, a journal whitelist, an agent that triages, and a Zotero database the agent writes to directly.

The single habit that made agents useful for my research work was not a better prompt. It was a folder of Markdown files that tell the agent where things are, what the hard rules are, and what has not been verified.

What happened when an agent rebuilt two of my oldest analyses from scratch, and what it taught me about using agents as reviewers of past work.

Ten tips from applying for data science roles and hiring data scientists, from showing business value to asking about growth.

Lessons from building a mobile game studio's first data pipelines: choosing tools, messy CSVs, metadata, error alerts and shared validation.

Query Redshift with psycopg2, shape the result in pandas, email it as an HTML table with smtplib, and schedule it with cron on a Mac.

The 2018 Lending Club scorecard rebuilt in Python with optbinning, an out-of-time test, a monotone LightGBM challenger and real reject inference.

How I handled my first year as the only data scientist at a start-up: research, networking, trial and error, and building confidence.

Two-way turnover rate grids on the Kaggle HR dataset, redone in pandas and checked against a boosted model with SHAP.

What a year of building risk scorecards at an alternative business lender taught me about domain knowledge, interpretability and operations.