AI Tech Talk @ SAP
Date: 27. 05. 2026
Time: 17.00
SAP, Holandská 2/4, 639 00 Brno
Another AI Tool Won't Fix This
Filip Prochac, Founder and CTO of Citymind
Everyone is looking for the next tool to boost productivity. But another tool will not fix what is broken. Most teams don’t struggle with technology. They struggle with how they approach their work.
In this talk, we will look at how to rethink the way we work. How to spot repetitive processes, break them into tasks, and start handing them off. Through practical examples, you will see why many efforts fall short and what it takes to make real progress.
Pattern Discovery at Scale Without Predefined Categories
Barbora Blaskova, Data Scientist at SAP
Every company with a support queue collects more tickets than any team can systematically analyze. The patterns (recurring failures, configuration gaps, undocumented edge cases) are buried in natural language, invisible without the right tools. We show how we surface those patterns without any domain knowledge or predefined categories using Large Language Models.
The talk walks through several techniques - from letting a language model build its own vocabulary, to shifting heavy processing to preprocessing so queries become instant, and covers what didn’t work first, what made it robust, and what transfers to domains beyond support data.
.
About speakers
Filip Prochac
Filip Prochac is the founder and CTO of Citymind, where he builds products used by cities in their day-to-day operations. Early on, his team built one of the first AI chatbots in the public sector, helping cities handle thousands of citizen requests, including deployments in Prague. That experience shifted his focus beyond communication to the work behind it. Today, he works on ways to simplify and gradually automate the processes cities deal with every day.
Barbora Blaskova
Barbora Blaskova is a Data Scientist in SAP, with over 4 years of industry experience and 3 years in academia. Her work in SAP focuses on applying AI and data analysis to derive insights from customer support cases.
Agenda
Everyone is looking for the next tool to boost productivity. But another tool will not fix what is broken. Most teams don’t struggle with technology. They struggle with how they approach their work.
In this talk, we will look at how to rethink the way we work. How to spot repetitive processes, break them into tasks, and start handing them off. Through practical examples, you will see why many efforts fall short and what it takes to make real progress.
Every company with a support queue collects more tickets than any team can systematically analyze. The patterns (recurring failures, configuration gaps, undocumented edge cases) are buried in natural language, invisible without the right tools. We show how we surface those patterns without any domain knowledge or predefined categories using Large Language Models.
The talk walks through several techniques - from letting a language model build its own vocabulary, to shifting heavy processing to preprocessing so queries become instant, and covers what didn’t work first, what made it robust, and what transfers to domains beyond support data.