- Actively investigating resource constrained deep learning algorithms for time-series data.- Team lead : Audio keyword detection framework- - Leading effort on redesigning framework for Python/C++ data generation and deep learning libraries to decrease training time and make them scalable.- - Optimized data processing pipelines for deep learning code-base, reducing training time by 50%.- - Benchmarked various CPU/GPU paradigms for keyword detection task taking into account file I/O bottlenecks and RAM constraints.- Alogrithm Lead: Multi microphone array user touch interface-- Designed and constructed initial proof of concept showing signal discriminability.-- Developed C/C++ library that provided linear/nonlinear SVM algorithms for capacitive-like touch interface on solid non-glass surfaces.-- Decreased training time by 70% and optimized algorithm robustness using MapR/Hadoop.-- Created Python framework for data collection, data analysis, and calibration of hardware.- Fault tolerant pedestrian localization system-- Implemented a multi-hypothesis tracking localization system in C,C++ that used asynchronous external measurements.-- Demonstrated algorithm on Android that used GPS and Bluetooth beacons.

- Productized and implemented fixed point navigation/orientation estimation/sensor calibration algorithms for deployment on embedded platforms in C.- Implemented C,C++ framework for data based orientation algorithm improvement.

I was a member of the core cross functional algorithm team that designed and delivered motion algorithms.- Developed a user/device rotation disambiguation algorithm for MEMS based indoor pedestrian navigation system in C,C++.- Implemented a decision tree stairs/elevator classifier using a MEMS pressure sensor in C,C++.- Increased robustness to highly dynamic magnetic/acceleration environments in an error-state extended Kalman filter by refining measurement models.- Performed root cause analysis of critical performance issues in MEMS based orientation and navigation algorithms.

* Adaptive Filtering: Developed a least squares filter with variable regularization that yields better performance than standard adaptive filters. Analyzed its convergence properties, computational complexity, and numerical stability.* System Identification: Developed a method that uses auxiliary/substitute nonlinearities for identifying systems whose inputs are corrupted by unknown nonlinearities.* Estimation: Developed an optimization based unknown input and state observer for nonminimum-phase systems. Analyzed the performance limits of state estimators for nonminimum-phase systems with unknown inputs.* Applied Data Assimilation: Applied a retrospective-cost-based input observer to identify the solar input and nitrous oxide cooling process in a full Earth atmosphere model using satellite data. Also applied this method to estimate the film growth on the electrode in a battery model.

Designed controllers for nonlinear systems with parametric uncertainties and input constraints using the Lyapunov direct stability theorem and log-barrier functions. Robustness was achieved by solving an optimization problem formulated in the framework of Polynomial Chaos expansion.

Empirically compared impedance pumping models, and designed an impedance pump in ANSYS/CFX. Fabricated an impedance pump to demonstrate the phenomenon.

Reviewed computer drafted designs pertaining to penstock, hydro turbines, digital governors and related ancillary equipment. Used quality control procedures related to fabrication, welding, radiography and installation of draft tubes, spiral cases, and steel penstock.

Optimized processes and material flow in pattern shop, fabrication shop, and machine shop.