ML Pub Club #23: GenAI Tradeoff - Train Small In-House or Prompt Large LLMs?
Details
Small models, in-house training… or massive LLMs at your fingertips? Which path gives you the best accuracy, speed, cost efficiency, and real-world reliability?
Join Patrik Mesec, Data Science Team Lead at Aircash, as he shares the story of their tiny vision-language model, the engine behind their automated identity verification system. From designing a model small enough to run efficiently in-house, to learning how to detect hallucinations and when to rely on large LLMs, Patrik will take you through the real-world challenges and decisions that shaped their AI journey.
Along the way, you’ll discover the pros and cons of in-house training vs. prompting large LLMs, how to compress models without losing performance, techniques to catch and prevent AI hallucinations.
🗓️ Tuesday I October 7th, 2025
🕖 6PM
📍 CroAI HQ's (Zavrtnica 17, 4th floor)
About Patrik:
With 4+ years of experience in fintech, computer vision, and automotive AI, Patrik has tackled projects spanning OCR, identity document recognition, face liveness detection, multilingual speech recognition, and more. A Kaggle Competitions Expert, he thrives on experimentation, collaboration, and turning AI breakthroughs into tangible business value.