Svetlana Timofeeva

Guideline Adherence: rigorously evaluated LLM outputs against detailed rubrics and safety policies, ensuring 100% compliance with complex task instructions. Structured Feedback: provided precise, actionable written feedback to training teams, identifying logical fallacies and factual inconsistencies in high volumes of data. Security & Policy: identified and flagged potential safety and integrity issues, contributing to the development of safer AI models for the Russian-speaking market.

Safety & Alignment: Conducted adversarial testing on large-scale Russian LLMs; investigated and mitigated technical abuse vectors and security vulnerabilities; analyzed datasets to detect coordinated model exploitation. Expert QA: Performed high-level reviews of model outputs focusing on linguistic accuracy, factuality, and ethical guardrails; reviewed sensitive, explicit, and high-risk content for safety and policy alignment. Adversarial Prompting: Developed complex, multi-turn prompts to stress-test robustness and jailbreaking resistance.

Led Russian-language quality assurance for AI model training projects, focusing on linguistic accuracy, cultural relevance, and overall localization of AI-generated content. Mastered a proprietary AI training platform to assess data quality, generate targeted feedback, and improve outputs for factuality, completeness, brevity, and grammar. Analyzed requirements and linguistic data to identify patterns, improve annotation consistency, and contribute to refining quality standards and guidelines for the Russian language. Wrote and refined prompt-based questions to produce diverse AI responses that meet high linguistic and stylistic standards. Collaborated with project managers and team members to clarify guidelines, address linguistic challenges, and optimize workflows, boosting team efficiency. Served as a subject-matter expert for Russian language and culture to ensure content appropriateness.