Code Club #4: Building an AI Harness: Harlequin

​As LLMs slowly enter the “as good as it gets” phase, the focus is shifting from the models themselves to the infrastructure around them - the systems that can actually make them useful in practice.

​In this session, we’ll talk about AI harnesses, memory models and the small architectural decisions that often end up having a huge impact on performance. How should context be structured? What kind of memory works best? Why can small changes in task formatting completely change the outcome?

​Through an experimental AI harness called Harlequin, we’ll explore approaches to memory management and context optimization, with a focus on smaller local models and the practical challenges that come with using them.

​🎤 Speaker:
Ivan Voras

​Voras is a researcher disguised as an IT architect and CTO, with more than three decades of experience spanning distributed systems, databases, digital electronics and, more recently, artificial intelligence.

​🗓️ June 17th
🕕 18:00
📍 CroAI HQ (Zavrtnica 17)

​🔗 Registrations via Luma.

​If you’re interested in what comes after the model itself and how serious AI systems are actually built around it, join us!