Nitish Kumar

Developed Deep Reinforcement Learning (DRL) algorithms with PyTorch, automating CAD-to-SEM image registration and eliminating manual calibration in semiconductor inspection, significantly increasing throughput.

Automated simulation parameter tuning using Optuna, resulting in a 40% reduction in manual calibration time and substantial improvements in on-site operational efficiency.

Engineered an automated ROI detection algorithm for 3D NAND metrology, utilizing signal and image processing to identify nitride band centers with 3 µm precision, reducing manual effort and error.

Implemented an ML-based classifier with Scikit-Learn and Python, achieving over 80% accuracy in automatically selecting optimal registration scores and differentiating successful from poor cases.

Enhanced coarse edge extraction accuracy by over 90% using Elastix and ITK-based registration algorithms in Python, critical for high-precision manufacturing metrology.

Prototyped a GAN-based image inpainting solution using PyTorch, significantly improving defect-repair performance and accelerating R&D project timelines.

Collaborated with international cross-functional teams across Israel, USA, and India to successfully deploy advanced AI/ML solutions in global facilities.

Integrated complex algorithms into production environments via CI/CD pipelines, adhering to Low-Level System Design principles for scalable and robust software delivery.

Engineered DSP-FPGA-based algorithms using Wavelet and C++ for triboelectric sensor intrusion detection, achieving 100% detection and over 90% classification accuracy.

Developed phase demodulation algorithms for fiber-optic PIDS using advanced Signal Processing in MATLAB, improving intrusion localization accuracy to within ±50 meters.

Applied Teager-Kaiser Energy Operator (TKEO) and Discrete Wavelet Transform (DWT) for robust feature extraction, integrating with ML models in Python for real-time signal analysis under strict computational constraints.

Built and validated Simulink models in MATLAB for high-voltage energizer systems, streamlining prototyping cycles and accelerating development by 30%.

Optimized embedded signal-processing code in C++ for low-memory, real-time execution, enhancing performance in critical defense-grade security applications.