What is One-Class SVM?
Machine Learning / AI
What LLM evaluation benchmarks do you know (MMLU, BBH, GSM8K, HumanEval)?
What are the strengths and limitations of SAM in production?
What is label smoothing in image classification?
What is 6D pose estimation of an object?
Как сравнивать стоимость инференса разных провайдеров и self-hosted решений?
When should hybrid search be used and how to weigh signals?
Tell me more about the anti-hallucination mechanism — what was meant by this mechanism in your system?
Tell us about QLoRA — what is it and how have you used it in your experience?
Tools for scoring credit applications — were they initially validated by analysts? How did the ReAct agent work with these tools?
Tell us about interesting projects in the field of classical ML — scoring, classification.
What types of augmentations are specific to detection (mosaic, mixup)?
What is query log mining and what signals does it extract?
What is the Cox proportional hazards model?
What features are there in training on multiple nodes (NIC, latency, jitter)?
How did you validate the AI agent — relying only on the customer's business metrics or using additional metrics?
What is OEC (overall evaluation criterion)?
What frameworks do you use (EconML, DoWhy, CausalML)?
What is the role of the text-encoder (CLIP) in Stable Diffusion?
How does function calling differ from traditional few-shot prompting for structured output?