MOPS - Meetup #4 [Poznań Edition]

Hi Pugaholics!

We are pleased to invite you for the first MOPS meetup in Poznań!

The formula is similar to previous editions three, about 30-minute practical talks followed by question-and-answer sessions with networking afterward.

The entire event and all talks will be held in English!

Key details:

Location: Allegro office, the ground floor of the D building of the Nowy Rynek (entrance from Wierzbięcice Street)

Insightful MLOps and LLMOps-related talks

Knowledge exchange during networking with pizza

Presentations:

• Building production-ready LLM software - We all know ChatGPT, but how do you from simple prompting to building a fully functional production app with LLMs? In this talk, I will discuss the various challenges you will face when building LLM software and how we can overcome them. We’ll dive into concepts such as RAG, observability, pipelining and more. By the end of this talk, you will have a good understanding of how to integrate LLM’s into traditional software development.
• A journey towards MLOps - from greenfield to a quite robust platform - I’ll describe how we built our ML Platform in Fandom from scratch. We started by training models in Jupyter Notebooks and deploying models from the given file - but now we reached much faster model delivery and deployment thanks to the automation we introduced. I’ll cover topics such as experiment tracking and model training pipelines and I’ll share the tooling we used to build our platform.
• Gotta go fast! A bag of tricks to speed up ML models inference in production - In many applications, especially real-time ones, the response time of the model is a critical factor. In some cases, if the inference time constraints are not met, it can make the whole pipeline impractical, or not even usable at all in the worst-case scenario. And even if there are no such constraints, having models run and provide predictions faster results in decreased computational requirements, leading to improved system scalability, energy efficiency, and cost savings. Having all those benefits should always sound good, but unless there are application-critical constraints, the inference optimization aspect is often overlooked. It shouldn't – as the technology evolved with time, there are now multiple tools that are low-hanging fruits that can be utilized, and often drastically speed up our models at almost no, or minimal implementation efforts. In this presentation, we will go through them, as well as the other more complex methods and tools discussing how to use them appropriately, and with what considerations – in order to make our ML models serving blazingly fast!

Speakers:

Łukasz Myśliński - A Co-founder at CvToBlind and MLnative, specialized in AI & MLOps. He was previously an Engineering Lead at Datarobot and Algorithmia, where he oversaw the development of large enterprise ETL and MLOps solutions. Originally with a Java/Scala background, he’s since turned his attention to building software in whatever gets the job done best, mostly with Python and Typescript. Privately a basketball & kitesurfing enthusiast.

Wojciech Mikołajczyk -

• Senior Machine Learning Engineer @ Fandom
• Experienced professional: Python -> Business Intelligence -> Data Science -> Machine Learning (5+ years) (https://www.linkedin.com/in/wojciech-mikolajczyk/)
• Mostly focused on NLP and MLOps
• Co-creator of ML-Workout (https://www.youtube.com/@ml-workout)
• Hobbies: Music, Piano (25 years+)

Michał Mikołajczak - A founder and Tech Lead at datarabbit.ai – data/machine learning focused software house, that helps organizations utilize their data and gain competitive advantage by designing, building, and shipping AI and data driven solutions for their businesses. Due to working there on a variety of projects from different industries, he possesses a broad range of diversified ML experience, with a focus on its productization. But his primary background is image processing – for a couple of years he was working in the medical imaging field, including being a CTO of a startup that was successfully acquired by a NASDAQ company. Privately a big fan of BI/any kind of data visualization systems that allow storytelling with data and Pratchett works enjoyer.