Posts
Notes from building analytics and LLM tooling that has to be right: NL2SQL, data modelling, and what breaks once it meets production data.
- How to efficiently push business teams to use AI
For AI leads who own adoption on top of everything else: start with LLMs in office tools, steer with org-level skills, run a champions program with real accountability, put a price on the hours saved, and keep the excitement from turning into duplicated work.
llms · ai adoption · leadership · 7 min
- How to be more ambitious with AI when working with dbt
Full test coverage, business-logic tests, column descriptions, tested exposures: the dbt work we used to skip because it took too long to type. With agentic engineering and a long voice prompt, you can do all of it, fast and under control.
dbt · llms · data quality · 6 min
- GPT-4o’s Nl2SQL Blind Spot: Inaccuracies on Sales KPI Calculations
One question, one simple table, three prompts: GPT-4o still cannot write a daily year-over-year sales query that holds up on real data.
nl2sql · llms · sales kpis · 7 min
- Challenges in Enterprise-Level LLM-Based Natural Language to SQL (NL2SQL)
Vanilla GPT-4, zero-shot, against a real e-commerce orders table. It got 15.6% of business questions right; here are the five ways the rest went wrong.
nl2sql · llms · sql · 6 min