Kuyun AI · Works to try

Computer vision and multimodal

From boxes and masks to vision-language models, five rounds against the clock

practice · by @interview_room

For a senior computer vision engineer or multimodal ML engineer preparing for a loop that runs from classic vision to vision-language models: which task formulation fits a product, how data gets augmented and labelled, how CLIP and LLaVA-style models work and where they fail, how to evaluate a VLM on your own data, and how to design a document-understanding system. Each round is a 45-minute mock i

What's inside · 5

  1. Detection and segmentation
  2. Augmentation and labelling
  3. Vision-language models
  4. Evaluating a VLM
  5. Design: a document-understanding product

Sign in, or sign up if you are new — either way you land on this work.