Gnaneswar Rao Gorle

Engineered and deployed ultra-fast face detection and 98-point facial landmark detection models tailored for real-time inference under constrained edge conditions. Developed a production-grade firmware simulation framework to monitor ML model drift and maintain inference integrity through periodic validation using reference datasets. Designed an early-onset drowsiness detection algorithm leveraging temporal EAR dynamics, blink rate decay, and PERCLOS to anticipate fatigue events before visual collapse. Implemented multi-signal driver behavior classifiers to disambiguate between downward gaze, visual occlusion, and sleep-induced microsleeps using eye aspect ratio and blink entropy. Built multi-modal distraction recognition pipelines for phone usage, seatbelt violation, and smoking detection by fusing facial geometry, motion vectors, and object context. Automated hyperparameter tuning using Optuna with multi-objective pruning (latency, F1-score), integrated into a modular PyTorch Lightning training pipeline. Performed model quantization, layer fusion, and ONNX graph surgery for real-time deployment on Intel OpenVINO, Kneron NPU (via NEF), and Ambarella CVFlow SDK platforms.

Built time-series DL models to forecast power utility consumption and detect anomalies. Designed a resume-screening system using NLP for job-resume matching and ranking. Delivered machine learning bootcamps for internal upskilling initiatives.

Developed a real/fake face recognition pipeline with person re-identification. Built a multimodal engagement detection model combining emotion (video, audio, text). Worked on EEG signal classification for left/right-hand movement recognition.