SURAJ PAUL

Engineered an end-to-end automated inventory inspection application for textile logistics. Trained custom object detection models capable of identifying overlapping fabric patterns and counting stored units accurately under challenging industrial lighting.

• Built an interactive Streamlit UI providing live feed visualization, dynamic threshold tuning, and exportable counts.
• Optimized bounding box non-maximum suppression (NMS) to eliminate multi-detection errors on stacked garments.

Developed an edge vision system tracking joint angles across heavy compounds (squats, bench press, deadlifts) to identify form degradation in real time.

• Audio-visual warnings fire immediately when anatomical angles deviate beyond safe mechanical thresholds.
• Equipped with kinematic review logs and rule-based hypertrophy workout and macronutrient suggestions.

Designed an access control infrastructure linking facial biometric authentication directly to physical solenoid locks over an encrypted bus.

• Validates user identity via FaceNet embeddings, triggering Telegram security alerts with image captures on intruders.
• Transmits cryptographically signed (ECC + AES-GCM) unlock pulses to prevent hardware interception.

Built a multi-target facial recognition attendance logging system running concurrently on commodity hardware with automated reporting.

• Handles simultaneous multi-student detection with temporal embedding averaging to minimize false rejections.
• Automates daily/weekly statistical attendance report generation transmitted directly to faculty Telegram channels.

Engineered a vehicular coprocessor monitoring operator alertness via Eye Aspect Ratio (EAR) metrics alongside forward-facing obstacle ranging.

• Triggers low-latency physical buzzer alarms when sustained eyelid closure or head drooping is detected.
• Integrates forward ultrasonic telemetry and object bounding boxes for proactive collision warnings.