Comer

Demonstrated experience with AI model evaluation.

Demonstrated experience with data annotation and labeling.

Demonstrated experience with RLHF.

Demonstrated experience with prompt engineering and evaluation.

Demonstrated experience with LLM assessment video/image annotation.

Focused on video and image annotation, primarily generating training data for generative AI models from video sources.. Labeled complex visual patterns, action sequences and object interactions across thousands of video frames while keeping labels temporally consistent.. Caught behavioral patterns and edge cases in footage, flagging anomalies likely to affect model performance.. Held 80%+ accuracy on video annotation tasks in Label box and Scale AI with 81.4% overall accuracy across the project's mixed media types.. Built diagnostic prompts and integration tools to strengthen AI module functionality and generation quality.. Worked with project leads to sharpen annotation rubrics and resolve labeling disagreements as they came up.. Maintained a steady 35–40-hour weekly schedule across the training initiative.

Wrote prompts, Q&A content, and text-to-video (T2V) generations to expand AI model capability across formats.. Produced image-to-prompt descriptions and assessed language quality in AI-generated responses for accuracy and context.. Used pattern recognition to catch recurring AI output errors and turned them into structured feedback for model refinement.. Ran side-by-side sample comparisons and rubric-based reviews to keep quality consistent across large batches.. Held a 80%+ accuracy across prompt generation and language evaluation while supporting two major clients.. Managed a high-volume 30-40-hour weekly schedule, adapting to shifting requirements and tight turnarounds.

Labeled human actions and behaviors in retail video footage, capturing how people interacted with merchandise, shelving and each other.. Applied action-boundary detection to mark precise start and end points of activities, helping the model learn temporal sequence.. Identified behaviors such as product selection, pin browsing, checkout, and social interaction within retail spaces.. Reviewed AI-generated footage and improved output quality by rewriting prompts for clarity and precision.. Answered structured Q&A on task footage to support model training and evaluation.. Kept a consistent 35–40-hour weekly schedule for the duration of the project.