Abdulahi Bello

Designed, trained, and evaluated machine learning models in Python (scikit-learn, pandas, NumPy) across 6+ projects spanning classification, regression, and clustering, improving baseline accuracy by up to 18%. Built and deployed 5+ model-serving REST APIs and lightweight web applications with Flask to expose trained models through documented endpoints. Engineered reusable data cleaning and preprocessing pipelines, reducing preparation time by ~30%. Curated and validated 15,000+ training/evaluation records and ran benchmarking loops against defined success metrics. Documented experiments and architecture to support reproducibility and smooth team handover.

Annotated 1,500+ videos and 12,000+ frames, labeling objects, actions, and events along timelines to support computer-vision and video-understanding training data. Created precise captions and transcriptions for 200+ hours of footage aligned to timestamps within ±0.2 seconds for subtitle generation and video-language model training. Segmented long-form video into labeled clips and annotated start/end times for key moments and scene changes with sustained throughput (~40 clips/day). Followed labeling guidelines across 10+ projects, flagged ambiguous cases, and improved dataset rules; performed QA on peer work maintaining ~98% acceptance rate and strong inter-annotator agreement.

Delivered 20,000+ image annotations across 8+ Meta projects, including bounding boxes, polygons, semantic segmentation, and keypoint/landmark labeling at high daily throughput (350+ images/day). Annotated 2,500+ short-form videos (Reels), labeling actions, objects, and content attributes frame by frame for content understanding and recommendation systems. Produced 3,000+ 3D annotations using SRT labeling to support depth, tracking, and scene-understanding models. Maintained ~98% accuracy against strict Meta data standards; performed QC audits to keep rejection rate below 2%.