YOLO brand detection for sponsor analytics
DosBro — YOLOv4 framework processing 300K+ social images/month for brand detection (89% mAP).
DosBro — YOLOv4 framework processing 300K+ social images/month for brand detection (89% mAP).
DosBro — containerized Deep SORT tracking for real-time brand exposure across 1M+ video frames/week.
Northeastern — BERT and ClinicalBERT on MIMIC-III for assessment-and-plan reasoning in clinical notes (Macro F1 0.78).
Northeastern — Hazard View Bird: EfficientNet-B0 segmentation for 14 disaster damage types, deployed on Jetson.
Northeastern — CNN framework on EEG spectrograms for detecting harmful brain activity (81.9% accuracy with EfficientNetB2).
DosBro — LSTM + XGBoost hybrid model for user engagement prediction (+15% accuracy vs. baseline).
NMIMS — VINVL-based system matching tweets to GIF replies from 1.5M tweets and 115K GIFs (80%+ accuracy).
NMIMS — ensemble classifiers on Framingham Heart Study data (85.2% accuracy); published at IEEE CONIT 2021.
DosBro — BERT/T5 summarization for JioTV companion apps (ROUGE-L 0.88, 45% faster editorial workflows).
DosBro — PyTorch QA system supporting dynamic content queries in three Indian languages (+30% engagement).
Pheme Software — LMS-integrated exam system with facial recognition proctoring (95% misconduct detection accuracy).
NMIMS — transfer learning on chest X-rays with custom preprocessing (97.4% recall); published at IEEE ICAIS 2021.