Uploaded February 2026 | Updated September 2026, 3 weeks ago
When a bug is fixed, how can we make sure it will not be re-introduced in the future?
Of course, through a proper regression test, i.e., a test that fails in the presence of the bug and passes in its absence.
However, regression tests are frequently ommited by the developers, who tend to submit the patch alone.
How can AI automation help developers it this scenario?
We explore this question in depth by introducing our open-source GitHub bot that is triggered when a pull request is opened, and proposes a test that is guaranteed to fail in the presence of the bug and pass in its absence.
We use a combination of LLMs and traditional, search-based testing techniques to achieve this goal.
The bot is currently deployed in three Python repositories (Javascript and Rust will be soon also supported) of the Mozilla corporation, where it has already helped developers with their regression tests. We share interesting early insights and feedback from this deployment and invite more open-source developers to adopt the bot and provide constructive feedback!
When a bug is fixed, how can we make sure it will not be re-introduced in the future?
Of course, through a proper regression test, i.e., a test that fails in the presence of the bug and passes in its absence.
However, regression tests are frequently ommited by the developers, who tend to submit the patch alone.
How can AI automation help developers it this scenario?
We explore this question in depth by introducing our open-source GitHub bot that is triggered when a pull request is opened, and proposes a test that is guaranteed to fail in the presence of the bug and pass in its absence.
We use a combination of LLMs and traditional, search-based testing techniques to achieve this goal.
The bot is currently deployed in three Python repositories (Javascript and Rust will be soon also supported) of the Mozilla corporation, where it has already helped developers with their regression tests. We share interesting early insights and feedback from this deployment and invite more open-source developers to adopt the bot and provide constructive feedback!
![[VDBUH2026] Brian Vermeer - Breaching LLM-Powered Apps: Overcoming Security and Privacy Challenges](https://i.ytimg.com/vi/iU0ytnOo6Qw/mqdefault.jpg "[VDBUH2026] Brian Vermeer - Breaching LLM-Powered Apps: Overcoming Security and Privacy Challenges
LLMs accessing the database and intelligent agents that perform online purchases? The possibilities for AI in applications seem endless but so are their security and data privacy risks. In this session, we’ll address common issues such as prompt injection, key leakage, abuse of private customer data for model training, legal restrictions, and more. In addition, we will show that general security issues in your systems can also influence the behavior and outcome of LLMs.
During this session, you’ll get a solid overview of the vulnerabilities to avoid, strategies to ensure data privacy compliance and best practices for building secure LLM-powered applications.")
![[VDBUH2026] Alexander Chatzizacharias: How to build your own fun and absurd pair programmer](https://i.ytimg.com/vi/jnPtYh-n0LM/mqdefault.jpg "[VDBUH2026] Alexander Chatzizacharias: How to build your own fun and absurd pair programmer
Tired of AI assistants that are always so boring and soulless? Alexander was. So, he decided to build his own. An AI assistant with personality, flair, and a healthy dose of sarcasm. Imagine a pair programmer that offers sarcastic feedback, makes absurd suggestions, and threatens to blow up your code when it disagrees with your changes. And when things get too quiet, it might even challenge you to a game of tic-tac-toe.
This session is for anyone who believes the best way to learn new technologies is by playfully breaking them. If you’re curious about LLMs and agents beyond the typical use cases, this talk is for you. You’ll leave with practical insights into building your own agentic LLM workflows using Spring Boot, vector databases, and locally running models. Alexander will talk about Retrieval-Augmented Generation (RAG) flows that feed LLMs the right context, multi-vector search for fast context retrieval, and Model Context Protocol (MCP) integrations that let the assistant directly meddle with your file system.
Come to learn, chuckle, and get inspired to create your own dysfunctional digital sidekick.")
