Deepak Kumar

About Deepak Kumar

Outlier benefited from my role as an Annotation & QA Specialist, where I ensured accurate datasets by focusing on data annotation, error detection, and validation across diverse formats. Collaborating with distributed teams, I contributed to improved workflows through guideline enhancements and tool usability suggestions. With a Master of Science in Mathematics from Lalit Narayan Mithila University, I combine technical expertise with a commitment to data precision. My work aligns with advancing AI-driven technologies through structured, reliable, and culturally relevant data annotation practices.

Experience

1. Image Rating Expert

   Abaka AI

   Jan 2026 — Present

   I specialize in the qualitative evaluation and refinement of Generative AI models. My work involves assessing synthetic media to ensure high-fidelity outputs, strict adherence to complex prompts, and visual consistency. I provide expert human feedback to help improve model alignment and reduce technical artifacts in AI-generated imagery.Key Responsibilities & ExpertiseGenerative AI Auditing: Conducted detailed assessments of model-generated outputs using standardized metrics, including Instruction Following, Visual Consistency, Technical Quality, and Naturalness.Comparative Output Ranking: Managed multi-response evaluation sets, applying relative ranking methodologies to determine the most accurate and high-quality results among various model completions.Identity & Anatomical Consistency: Specialized in identifying subtle deviations in human anatomy, facial features, and structural proportions to ensure subject identity remains consistent across edited versions.Technical Artifact Detection: Analyzed and documented visual anomalies such as lighting inconsistencies, depth-of-field errors, and unnatural object placement to improve model realism.RLHF Pipeline Contribution: Provided structured qualitative feedback and data-driven justifications to support Reinforcement Learning from Human Feedback, directly assisting in the fine-tuning of Large Language and Vision Models.Guidelines Compliance: Strictly adhered to evolving technical specifications and linguistic requirements to maintain high data integrity for model training sets.

Education

• B. P High school

  Intermediate, Mathematics

• Lalit Narayan Mithila University, Darbhanga

  Master of Science - MS, Mathematics

• Lalit Narayan Mithila University, Darbhanga

  Bachelor of Science - BS, Mathematics