PyData Chicago

Talk 1: Repeatable repeats with rpeat ® - DevOps for Science by Jeff Ryan

Details: When cron is not enough, and everything else is too much. This talk will introduce a new open source scheduler that is language neutral, highly scalable, easy to use, and has zero dependencies. Written in go if you care, but usable for everything. Run in production internally and by others.

Audience Takeaway: Designed for everything from small projects to large scale compute, and anywhere in between, you can now easily run and manage jobs with the simplicity of cron, but with all the power or the most widely known tools like AirFlow or Prefect. The talk focuses on using the software, but can dig into OSS licensing, design and implementation if there is interest.

Talk 2: Predicting Software Vulnerabilities: From Code Metrics to LLM-based Detection by Puya Pakshad

Details: In this talk, I will share how we can predict software vulnerabilities using both traditional code metrics and newer methods like large language models. I will start by presenting the core idea from a framework I published in 2023, which uses simple static analysis features and machine learning to identify vulnerable functions in C/C++ code. Then, I’ll introduce a newer approach I’m currently working on that uses LLMs to detect vulnerabilities directly from the source code. It’s still under review, but I’ll show how it compares to the older method in terms of accuracy and scalability.

I’ll also talk about the bigger picture in vulnerability detection research, what kinds of tools are being developed, and how automation is playing a growing role. The examples I’ll share come from real projects, and I hope the talk gives practical insight into how these ideas can be used in actual development work.

Audience Takeaway: What the audience can expect from this talk is a practical understanding of how software vulnerabilities can be predicted using both traditional methods based on code metrics and modern techniques powered by large language models (LLMs). They will explore the strengths and limitations of each approach, review real-world examples and experimental tools, and learn how incorporating LLMs into vulnerability detection can significantly improve accuracy, effectiveness, and scalability.