AI based proofs of concept are popping up everywhere. But taking an AI solution from prototype to production is about more than just solving technical challenges—it's about earning trust.

AI systems are often perceived as complex black boxes, which can make it difficult for users to fully understand or feel confident relying on them. In this edition of AI Talks, we explore what it takes to build AI systems that people actually want to rely on, covering topics such as adoption, evaluation, and uncertainty estimation.

📅 Date: May 21st, 2025
🕕 Time: 6:00 PM onwards with a drink (we'll start at 6.30 PM)
📍 Location: Cantersteen 47, Central Gate, 1000 Brussels (at Superlinear's office)

Talk #1: Evaluating LLM-based Applications – From Principles to Practice

LLMs can unlock powerful workflows, but how do you know if they’re actually working and don’t regress over time? In this talk, we break down the core principles of LLM app evaluation, from building test sets and choosing the right metrics to structuring a full evaluation pipeline. Then we zoom into a real-world use case at Exalate, showing how we evaluate our assistant during development and monitor it in production. Whether you’re building your first LLM app or scaling one in production, this session offers practical advice for making evaluation part of your process.

By Usman Shani (AI Engineer at Exalate) and Stijn Goossens (Solution Architect at Superlinear)

Talk #2: From Pixels to Impact – Automating Car Damage Detection with Vision AI

Macadam inspects thousands of vehicles every day. Ensuring consistency, speed, and accuracy across inspections is a massive operational challenge. In this session, we share how we used computer vision to automatically detect damages—turning a manual process into an AI supported workflow. We’ll walk through how we co-designed the solution with users in the loop and what it took to build trust in the application throughout the organisation as we moved from prototype to production.

By Dieter Maes (Product Owner at Macadam) and Mathieu Kitenge (Machine Learning Engineer at Superlinear)

Talk #3: Taming uncertainty with conformal prediction

Machine learning is increasingly used in high-stakes decision making—from predicting electricity demand and stocking shelves to diagnosing medical conditions. But how do we know whether to trust an AI’s prediction? In this talk, we’ll introduce conformal prediction, a powerful method for quantifying uncertainty, and show how the open-source package, Conformal Tights, lets you move beyond point predictions to generate reliable prediction intervals that guide better, safer decisions.

By Jan Wuzyk (Solution Architect at Superlinear)

Why join us?

At the AI Talks by Superlinear, you’ll…

• Network with top AI professionals from Belgium

• Dive into practical AI applications shaping industries

• Enjoy great finger food, drinks, and inspiring conversations