BrowserStack Online Meetup Stockholm

Testing the Untestable: Lessons from AI Evaluation

BrowserStack Tech-Leader Series | Virtual Meetup

Traditional software testing operates on a comforting premise: if you provide input $X$, you will always receive predictable output $Y$, verified by clear, binary pass/fail criteria. Generative AI has shattered that premise entirely.

When dealing with Large Language Models (LLMs), outputs are probabilistic, fluid, and context-dependent. How do you construct an automation framework for a system designed to never give the exact same answer twice? How do you catch subtle hallucinations, biases, or systemic quality drops before they hit production?

In this session, we strip away the theory to look at the raw mechanics of AI evaluation. We will explore how traditional QA philosophies are adapting to handle non-deterministic systems, moving from rigid code assertions to dynamic data validation, semantic evaluation, and targeted automated oversight.

🔍 Why This Session Matters

The Death of Binary Validation: Transitioning away from classic assertions toward validation methodologies suited for unpredictable, fluid model responses.

LLM Evaluation at Global Scale: Real-world lessons from auditing, refining, and grading millions of natural language processing (NLP) data points for enterprise-grade AI models.

The Evolving QA Skillset: How mathematical analysis, data profiling, and statistical validation are becoming core pillars of the modern quality engineering toolkit.

💡 Key Takeaways

Master the metrics used to benchmark and evaluate AI model readiness before deployment.

Bridge the gap between classic test automation principles and modern data-centric AI quality assurance.

Design fallback systems and human-in-the-loop oversight frameworks that keep probabilistic software under absolute control.

🎙️ About the Speaker

Jasna Bogunović Marković

Data & AI Quality Analyst & STEM Expert | Invisible Technologies

Jasna brings a unique, highly analytical perspective to the world of quality engineering, combining over two decades of advanced mathematics and statistical analysis expertise with cutting-edge AI data validation. Based in Serbia, she serves as an Advanced AI Data Trainer and QA Lead at Invisible Technologies, where she orchestrates comprehensive error analyses, refines complex NLP training data, and validates large language model performance prior to deployment for global technology enterprises. With a deep foundation in academic leadership, mathematics, and data analytics, Jasna specializes in translating chaotic, non-deterministic system behaviors into structured, measurable quality standards.

📅 Event Logistics

Format: Online / Virtual Meetup

Target Audience: Stockholm Tech Community (Open to global registrations)

Date: Tuesday, June 30th, 2026

Time: 6:00 PM Stockholm CEST