Data Operations IL: Managing Data for Vision Models
What makes computer vision such a powerful yet operationally demanding domain?
Between massive datasets, complex annotation pipelines, and the need for constant monitoring, Computer Vision Operations is where AI meets reality - and where every small detail can make or break your model’s performance.
Data Operations IL & Imagen have joined forces for this event to explore the world of Computer Vision Operations - where raw image data turns into structured intelligence, and production systems must deliver accuracy, speed and reliability at scale.
Computer vision may look like a solved problem in demos - but under the hood, it’s a continuous cycle of data curation, labeling, retraining, validation, and governance. In this meetup, we’ll hear real-world perspectives from teams that operate large-scale CV systems, and learn how they keep their models robust in production.
We’ll hear two talks from practitioners at the forefront of the field, followed by Umm, Actually - an interactive data game show that will put your data intuition (and sense of humor) to the test.
Speakers:
Maria Yatskevich, AI Operations Manager @ Imagen
Maria will share how rethinking the way AI training data is collected can transform results. This case study shows how operational decisions - from workflow design to managing contributors with the right domain expertise, can turn a challenging data task into a high-quality, scalable process that directly improves model performance.
Katya Alfandari, Data Engineer @ Blink
Katya will examine the challenges of processing and labeling facial features for eye-movement inference, where small variations in annotation or model interpretation can produce markedly different results.
Umm, Actually: Data Operations Computer Vision Edition
Hosted by Tal Mizrachi (Analysis Paralysis) and Dean Langsam (Data Scientist @ SentinelOne)
From mislabeled bounding boxes to false positives that slipped through QA - in the world of computer vision, even the smallest “actually…” can spark a debate.
In this live game show inspired by the cult-favorite Umm, Actually, contestants (and the audience!) will face statements about data, data operations, and computer vision — some true, some almost true - and compete to spot the inaccuracies first.
Schedule:
18:00 Mingling, Snacks & Free LinkedIn Photoshoots provided by Imagen
18:30 Opening Words
18:40 Talks
19:20 Umm, Actually Data Game Show
20:00 Wrap-up & Networking
* Event will be in Hebrew, and will be recorded