Avinash Thatavarthi

Annotated and labeled complex datasets across text, code, and structured data formats according to detailed project guidelines; reviewed 500+ annotation examples per sprint; maintained annotation consistency rates above 96% across multiple data streams contributing to LLM training and evaluation data quality. Evaluated and rated AI-generated responses for quality, accuracy, relevance, and safety by designing and operating custom eval harnesses and LLM-as-judge scoring pipelines that covered 500+ domain-specific test scenarios; provided structured quality ratings and actionable feedback to inform model training, prompt engineering, and guardrail improvements. Moderated AI-generated content to ensure policy compliance by applying content safety policies, identifying harmful or biased outputs, and flagging edge cases with detailed feedback for remediation teams, improving content safety guardrails over time. Delivered consistent annotation and evaluation output within deadlines by managin

Annotated and labeled complex training datasets across text, structured data, and code formats for enterprise AI model development; maintained annotation consistency rates above 94% across batches and contributed to AI model quality improvements. Evaluated AI-generated responses via Python-based evaluation pipelines with LLM-as-judge scoring and human review queues, driving 18% improvement in model accuracy. Moderated AI-generated content for policy compliance and safety, producing detailed moderation reports with specific policy violation citations and remediation actions. Independently managed annotation task queues across multiple concurrent projects, delivering all batches on time with quality above team averages. Built Python automation scripts to support large-scale annotation consistency analysis, improving throughput by 40%. Proactively communicated project status and blockers in a fully remote setup.

Performed data annotation and labeling for AI model training datasets across text, structured data, and code formats, ensuring high consistency and meeting quality standards for downstream model training. Evaluated AI-generated outputs, identifying errors, inconsistencies, policy violations, and providing structured ratings and explanations. Reviewed and moderated AI-generated content for policy compliance and safety, escalating recurring issue patterns to refine guidelines. Built Python scripts to automate repetitive data preprocessing and annotation consistency checks; delivered annotations on schedule. Communicated feedback and issues clearly through structured written reports and adapted to evolving guidelines while working remotely.