Free Workshop on Agentic Analytics for Data Scientists
Agentic Analytics in Production
Most data scientists are learning agentic AI. Very few are shipping it. This workshop is the head-start I wish I'd had.
Plus a 7-day email course: AI as a system, not a tool.
Agentic Analytics Workshop
30-minute deep dive
What you'll walk away with
You've read the tutorials. You've seen the demos. The gap isn't knowledge, it's architecture. This workshop walks through how production agentic analytics actually comes together, which trade-offs matter, and what to build first when you sit down to design your own.
See where tutorials end and production begins
The specific failure modes when real data, real users, and real edge cases hit. The places where most courses cut to a working demo and skip the parts that matter.
Learn the four pieces that decide whether agentic analytics works in production
Semantic models, guardrails, routing logic, and observability. What each one does, where each one breaks, and why every demo skips at least three of them.
Watch a real production system come together end to end
A walkthrough of a Talk-to-Your-Data Slackbot I shipped in production, from architecture to deployment. Trade-offs visible. Not a toy demo.
Walk away with the intuition to design one yourself
By the end, you'll know how the pieces fit together, where the trade-offs are, and what to build first when you sit down to do this in your own work.
Who's teaching this
Andres Vourakis
Senior Data Scientist with 8+ years in tech and applied AI/ML. Shipped a Talk-to-Your-Data Slackbot in production at my current company, an agentic analytics system built on the patterns covered in this workshop. Presented this workshop at DataFest and other data and AI events. The workshop on this page is the version of that talk I wish I'd had when I started.
What is agentic analytics, and why data scientists specifically?
Agentic analytics is the practice of building AI systems that reason over data autonomously: routing requests to the right sources, fetching context, validating their own assumptions, and returning reliable answers humans can act on. For data scientists, it sits at a specific point in the field: where understanding agentic AI stops being enough and shipping a working system becomes the actual job. Talk-to-your-data Slackbots, automated EDA agents, and multi-step analysis workflows are common examples.
Why this is a data scientist problem
Agentic analytics sits at the intersection of system design and domain knowledge. The semantic layer, the evaluation criteria, and the trust calibration are all things only someone close to the data can build correctly. Throwing these systems over the wall to an AI engineering team produces something that runs but doesn't know what a good answer looks like. Tools like LangGraph, MCP servers, and evaluation harnesses are useful, but they don't replace the judgment that comes from having lived in the data.
What "production-ready" requires
Production-ready agentic analytics means more than "it runs in a notebook." Concretely, the bar is:
- →Semantic models the agent can reason over without hallucinating columns or joins.
- →Context engineering that keeps prompts predictable across questions and users.
- →Guardrails and question-quality rubrics so bad inputs are caught before they hit the data.
- →Routing logic that directs each request to the correct data source.
- →Observability on every output, so you can debug when the agent misbehaves.
- →Evaluation against real business questions, not vibe-checks.
- →Graceful failure modes for the questions the agent shouldn't try to answer.
This is what the workshop covers. It's also what most tutorials skip.
How agentic analytics differs from things you've already heard of
Agentic analytics vs. text-to-SQL
Text-to-SQL turns one question into one query. Agentic analytics is a system: it routes the question, fetches context, runs the query, validates the result, and returns reasoning alongside the answer. SQL generation is one step in it, not the whole thing.
Agentic analytics vs. RAG
RAG retrieves documents and lets a model answer over them. Agentic analytics goes further. It reasons over structured data, routes between tools, and takes multi-step actions to produce an answer. RAG can be a component of an agentic analytics system, but it isn't the system.
Agentic analytics vs. AI engineering
AI engineering is the broader discipline of building production AI systems. Agentic analytics is the specific application of those skills to data science work: semantic layers grounded in your team's data, evaluation against actual business questions, and trust calibrated to your stakeholders. The data scientist owns the parts an AI engineer can't.
Take the head-start. Plus the 7-day course.