Agentic Analytics in Production: Lesons from Building a Talk-to-your-data Slackbot
Who is this talk for?
This talk is for data scientists and analytics professionals who want to understand where the field is headed and develop the practical AI skills that will matter most in the coming years.
Overview
Agentic analytics is becoming one of the defining shifts in 2026, but building these systems in production requires far more than connecting an LLM to a data warehouse.
In this session, I break down the architecture behind the Talk-to-Your-Data Slackbot I deployed at Nextory, an AI system that allows stakeholders to access insights conversationally on demand, freeing data scientists to focus on more complex work.
You’ll learn how:
Semantic models enable context-aware reasoning
Guardrails and question-quality rubrics ensure clean, reliable inputs
Routing logic directs each request to the correct data source
Observability and constraints keep agent behavior predictable in production
I’ll also cover the engineering work teams often overlook, including UX, adoption, monitoring, and the trade-offs required to make agentic systems usable beyond demos.
📌 Expect a highly practical, engineering-focused deep dive into what it actually takes to bring agentic analytics into production.
By the end, you should have a strong intuition for how to design and build one yourself.
Structure
⏱️ 45 mins: The talk
⏱️ 15-20 mins: Live Q&A
About me
My name is Andres Vourakis, and I'm a Senior Data Scientist at Nextory, where I’ve spent the last year integrating AI directly into day-to-day workflows. I've also built an LLM-powered product used by hundreds of users.
This talk is based on a workshop I gave at DataFest last November, which unfortunately was not recorded, and this webinar is the first time I’m sharing this material publicly.