Ashwin Kamble

<p>Developed ML pipelines for DFS event detection in radar-based wireless systems using large-scale real-world RF datasets across multiple regulatory regions.<br>
Evaluated and optimized ML models based on accuracy, latency, model size, false alarm rates, and deployment complexity for embedded real-time environments.<br>
Implemented optimized inference pipelines in C for internal testing and system-level validation.<br>
Built LSTM-based models for packet/preamble detection in WLAN signals (12.5k waveforms, 3 GB dataset), significantly reducing false detection under noisy and impaired conditions.<br>
Improved RF signal classification through feature engineering, preprocessing, and class imbalance.<br>
Explored LISTA-based approaches for channel coding and sparse signal recovery.</p>

<p>Developed ML pipelines for DFS event detection in radar-based wireless systems using large-scale real-world RF datasets across multiple regulatory regions., Evaluated and optimized ML models based on accuracy, latency, model size, false alarm rates, and deployment complexity for embedded real-time environments., Implemented optimized inference pipelines in C for internal testing and system-level validation., Built LSTM-based models for packet/preamble detection in WLAN signals (12.5k waveforms, 3 GB dataset), significantly reducing false detection under noisy and impaired conditions., Improved RF signal classification through feature engineering, preprocessing, and class imbalance., Explored LISTA-based approaches for channel coding and sparse signal recovery.</p>

<p>Developed signal processing and ML pipelines for EEG-based imagined speech classification using 50+ GB of multi-source EEG datasets and custom-acquired data.<br>
Designed customized CNN and used pretrained deep learning models for robust classification of noisy, non-stationary EEG signals in BCI communication systems.<br>
Applied wavelets, FFT, time-frequency analysis, adaptive decomposition, filtering, and ICA for feature extraction, noise reduction, and subject-independent classification.<br>
Built EEG acquisition, preprocessing, and real-time inference pipelines using MATLAB, Python, PyTorch, and TensorFlow.<br>
Published 4 SCI-indexed journal papers, 3 book chapters, and 1 conference paper.</p>

<p>Designed signal processing and CNN-based ML pipelines for arterial pulse analysis and cardiovascular risk prediction using data from 1200+ participants across hospitals and medical camps.<br>
Improved cardiovascular prediction performance by 20% through optimized preprocessing, feature extraction, and deep learning workflows, enhancing sensitivity and specificity.<br>
Developed AWS-based analytics pipelines and real-time algorithms for arterial pulse analysis, arrhythmia detection, and non-invasive cardiovascular assessment, improving diagnostic performance.<br>
Contributed to development of an AVF maturation monitoring device for nephrology applications, enabling real-time patient monitoring and clinician-facing insights.<br>
Supported system integration, validation, and deployment in clinical environments.</p>

<p>Designed signal processing and CNN-based ML pipelines for arterial pulse analysis and cardiovascular risk prediction using data from 1200+ participants across hospitals and medical camps., Improved cardiovascular prediction performance by 20% through optimized preprocessing, feature extraction, and deep learning workflows, enhancing sensitivity and specificity., Developed AWS-based analytics pipelines and real-time algorithms for arterial pulse analysis, arrhythmia detection, and non-invasive cardiovascular assessment, improving diagnostic performance., Contributed to development of an AVF maturation monitoring device for nephrology applications, enabling real-time patient monitoring and clinician-facing insights., Supported system integration, validation, and deployment in clinical environments.</p>