The YandexART team develops generative models for creating and editing images. Work on the agent capabilities of models: reasoning chains, using web search and other tools to solve multi-step tasks.
Responsibilities:
- Evaluating the quality of agent scenarios: developing metrics and quality measurements for complex image generation and editing (focus on reasoning chains, web search, VLM-feedback loop).
- Creating data generation pipelines for agent scenarios: generating reasoning trajectories, calling additional tools, labeling through judge models.
- Analyzing and improving training datasets: identifying growth areas, defining data quality metrics, analyzing the impact on generation quality, designing data collection and quality control of labeling.
Requirements:
- More than 1 year of experience in analytics.
- Proficient in Python and SQL for data analysis.
- Understanding of ML fundamentals and model quality metrics.
Conditions and Bonuses:
- Office-based work in Moscow.
- Large offices with gyms (exercise equipment, inventory, showers) and other bonuses.
