Developed an end-to-end simulation framework using Convolutional-Recurrent Neural Networks, achieving 97% accuracy in predicting voltage waveforms across 10+ Snapdragon generations. Reduced prediction latency by 90%, saving hundreds of man-hours.
Engineered an RL-based routing methodology for PCBs, achieving 50% cost savings on simulation tools and reducing development cycle time by 20%. Ensured stable output voltage levels and lower impedance factors.
Architected a unified platform integrating 5 disparate tools, streamlining workflow for monitoring current transients, moving averages, and current scaling. Currently used by 10+ R&D engineers.
Formulated lightweight transformer architectures for multimodal sentiment analysis, reducing processing time from 2-3 hours to 10 minutes while maintaining superior learning capabilities.
Presented efficient Transformer-based architectures for multimodal sentiment analysis, outperforming larger state-of-the-art systems.
Proposed an AI architecture combining neural networks and regressor trees to predict PDN characteristics, reducing evaluation time from weeks to minutes with 94% accuracy.
Designed a blockchain-based system for protecting image copyrights using perceptual hashes and Ethereum smart contracts.
Developed an attention-based neural network framework for Android malware detection, achieving 96.75% accuracy in behavior analysis.
Developed an approach using Gradient Boosting and Gaussian Processes to estimate traffic accident severity using spatial-temporal features.
Developed an automated system for dermatological disease recognition using CNN and SVM, achieving 95.3% accuracy in lesion detection.