Two Talks: Diving into the Heart of LLMs & Fundamental Patterns in Agentic AI Applications

Talk 1: Diving into the Heart of LLMs

Large Language Models are one of the fundamental pieces of the modern AI ecosystem - but hardly anyone understands how they actually work. It's a collective challenge, as it's very short sighted to rely your business and architecture on a technology that one might have a hardly any knowledge about.

And it is critical to build a proper understanding, as many good and bad ideas can be evaluated, thanks to seeing how the internals work. At least to some degree.

In this talk we dive into the heart of LLMs - from user's prompt, though tokenizers, embeddings, meeting the core - transformers - and finally output generation.


Talk 2: Fundamental Patterns in Agentic AI Applications

Agentic applications are just applications, though the underlying "service" they're using is somehow specific. There are specific database-related patterns, queue-related patterns etc - not to mention various challenges specific to each. And LLM-based systems are not different!

Both scientists and the industry have already identified some key advantages and issues in practical work with large language models. And hence, there's a handful of very practical patterns that help us overcome the limitations of the context window, issues with the quality of models and their high infrastructure costs. This talk is a distilled SOTA when it comes to practical Agentic Architecture.

Tomasz Ducin - Independent Consultant, Architect, Developer & Trainer, Poland