Somya Gurjar

• Strengthened dataset reliability by systematically auditing LLM pre-annotations, correcting false positives and missed events, and enforcing consistent labeling decisions across edge cases • Worked on pairwise evaluation, object removal and bounding boxes • Audit and refine LLM pre-annotations by removing false positives, adding missed events, and tightening labels • Execute time-aligned tagging with strong precision, ensuring reliability of sequence-based data signals • Write structured reasoning for non-obvious decisions, grounded in rubric definitions and classification rules • Contribute to calibration cycles and QA reviews to reduce drift and enforce dataset consistency

• Delivered high-precision annotation outputs under strict rule-based frameworks across varied datasets • Audited AI/LLM-generated labels, systematically correcting misclassifications and tightening decision boundaries • Enforced consistent labeling across edge cases through explicit rule application and structured reasoning • Logged ambiguous scenarios with clear assumptions, improving repeatability and downstream QA alignment