Senior Consultant - AI Training & Evaluation (MBB & Top-Tier Firms)

Mindrift, powered by Toloka, is a leading enterprise AI and machine learning data partner since 2014 that connects top domain experts with cutting-edge AI initiatives. The company supports frontier model post-training by building domain-specific reinforcement learning environments, tasks, and evaluation frameworks designed by real practitioners. Mindrift is launching a Management Consulting domain focused on translating real-world consulting engagements into structured learning environments for advanced AI systems.

• 3+ years of experience at McKinsey, BCG, Bain, Oliver Wyman, Roland Berger, Monitor Deloitte, EY-Parthenon, Kearney, or Strategy&.
• Strong structured problem-solving and hypothesis-driven thinking.
• Ability to translate vague problems into clear analytical steps and deliverables.
• High attention to logical consistency and output quality.
• Independent, self-directed working style.
• Clear written English proficiency at B2 level or higher.

• Build realistic consulting project environments with detailed project scenarios grounded in real engagement dynamics including industry context, financials, constraints, conflicting inputs, and incomplete information.
• Design structured consulting tasks for AI agents that mirror real consulting work such as market sizing, commercial due diligence, cost optimization, growth strategy, operational diagnosis, and benchmarking.
• Define evaluation criteria and quality standards by developing grading frameworks, evaluation rubrics, and golden-answer solutions for each task to train and calibrate an LLM-based grading system that evaluates AI outputs at scale.

This role is exclusively for consultants with direct experience at a top-tier strategy consulting firm: McKinsey & Company, Boston Consulting Group (BCG), Bain & Company, Oliver Wyman, Roland Berger, Monitor Deloitte (Deloitte S&C), EY-Parthenon, Kearney, and Strategy& (PwC). Candidates without hands-on project experience at these firms should not apply.