Ankur Roy
AI Data Specialist & Technical Evaluator
Hire this person
Send a job offer directly to this candidate
About
Detail-oriented AI Data Specialist and Technical Evaluator with proven production experience across Outlier, Remotasks, and the Atlas Data Annotation platform. Specializes in Reinforcement Learning from Human Feedback (RLHF), multi-turn reasoning critique, complex instruction following, and rigorous factuality verification. Combines engineering analytical rigor with an exceptional eye for detail to train next-generation generative AI foundation models.
Experience
AI Trainer & Evaluation Specialist at Outlier.ai (2024-01 – Present)
- Evaluate, rank, and rate multi-turn LLM candidate responses across truthfulness, adherence to multi-constraint instructions, conciseness, and tone.
- Author comprehensive justification rationales diagnosing subtle logical fallacies, edge-case rubric violations, and mathematical inaccuracies.
- Conduct source-level verification and fact-checking to detect and mitigate subtle hallucinations across STEM and reasoning tasks.
- Sustain a 95%+ quality audit rating across priority calibration queues and specialized project benchmarks.
Data Annotation Specialist at Atlas Data Annotation Project (2023-01 – 2024-12)
- Executed fine-grained intent classification, conversational context auditing, and multi-turn dialogue labeling to refine NLU training sets.
- Evaluated step-by-step reasoning (Chain-of-Thought) outputs, flagging ungrounded inferential leaps or subtle contradictions.
- Partnered asynchronously with distributed QA reviewers to harmonize edge-case interpretations and standardize scoring rubrics.
AI Data Contributor & Evaluator at Remotasks (2023-01 – 2024-12)
- Mastered extensive 20+ page dynamic guidelines to deliver high-throughput, high-accuracy text evaluation and taxonomy classification.
- Identified prompt ambiguities and guideline conflicts, actively providing constructive feedback to project leads to refine edge-case rules.
- Performed side-by-side preference evaluations to align AI outputs with human safety, helpfulness, and harmlessness policies.
Education
B.Tech in Civil Engineering – Heritage Institute of Technology (2025-06)



