Tech talk · January 2026 · 40 min
Plug and Play Data Analytics, the SQL-to-insight pipeline.
How to answer stakeholders' ad-hoc data questions with LangChain and PostgreSQL, without letting an LLM anywhere near your production database.
Big Data community tech talk
Full recording · 48 min
The problem
It's 4 pm and a stakeholder needs a "quick data pull". On the surface it's one SQL query. In practice it's a context switch: connect, check the joins, clean the CSV.
Dashboards don't fix it. They're static, so the first new filter sends the request straight back to your inbox. Data engineers end up as human SQL translators instead of building pipelines.
For
Data engineers and analytics professionals
Stack
What the talk covers
- 01
The risk
Why pointing an LLM at a production PostgreSQL instance is an operational hazard: unindexed joins on large tables, and hallucinated joins on a schema it doesn't understand.
- 02
OLTP vs OLAP
Transactional databases are built for writes, not questions. Analytical questions belong on a copy, an analytical reflection, not on the row-store production replica.
- 03
Schema as context
Giving LangChain the schema (tables, columns, types from information_schema) so the SQL it writes is grounded in your real model.
- 04
Medallion for AI
Bronze, silver and gold layers, so the model reads clean, named, documented tables instead of guessing at raw ones.
- 05
Two live demos
The Northwind e-commerce database, and a copy of a real web analytics (Umami) database answering questions like "What is the top visited page?".
Key takeaways
- —Context is king: the answer is only as good as the schema and data the model is given.
- —Never let an LLM query production OLTP. Query a copy built for analytics.
- —RAG isn't an AI feature; it's a retrieval optimisation problem, which is a data engineering problem.
- —With the plumbing right, the work moves up a level: from writing queries to understanding the business logic.
Want this in your stack?
A 30-minute decision call turns this pattern into a plan for your team, or book the talk for them.