Exploring the Future of AI & LLMs with Wix, Microsoft and IBM

This meetup focused on how to make LLMs actually work in real-world, enterprise environments - not just in demos. We saw how Wix customized models like LLaMA to their own data, even outperforming GPT-3.5 on specific internal tasks like QA and summarization - proving that smaller, tailored models can be more accurate, cheaper, and easier to control. From there, we explored how to constrain model outputs using tools like Guidance, enabling precise generation (like valid JSON or fixed-option outputs) instead of "free-form guessing." Finally, we learned how prompt engineering itself is evolving, with approaches like Conversational Prompt Engineering (CPE) turning prompt creation into an interactive, user-friendly process. The takeaway: building reliable AI systems isn’t just about the model - it’s about customization, control, and better ways for humans to communicate intent.