Developing Machine Learning / Data Science projects in the text and time-series domain to make SRE processes more efficient. Working exclusively in python. Projects:- Advancing observability with Root-Cause Analysis: Utilizing graph machine learning algorithms (GraphSAGE Neural Net), with Transformer encoder/decoder models (BERT, GPT) as well as unsupervised text clustering techniques (KMeans, HDBSCAN, UMAP), to develop SRE incident root-cause analysis applications. - RAG Chatbot: In-house chatbot end-to-end chatbot application for JPMC internal data/systems using the Retrieval Augmented Generation (RAG) pipeline. Responsibilities include creating data pipelines, developing fine-tuned Retriever and Re-Ranker, as well as validation testing for the LLM Reader. - AWS Simulations framework: Developed an AWS simulation framework for mocking realistic behavior of several AWS services, for unit testing and application simulation in JPMC.

Deep Learning for Gorilla Glass: Developed and deployed a custom ML solution, using Convolutional Neural Networks (CNN), for defect classification on 20+ Gorilla Glass inspection lines. Acascaded ML system is developed, with classical image processing techniques employed to catchcritical corner cases. ML Modeling is done using the Tensorflow-Python API, and deploymentdone by building a C# wrapper on the model files to communicate with the vendor equipmentusing TCP/IP protocols.Machine Health Estimation for Optical Fiber : Developed an ML Predictive Maintenance systemto estimate machine health and forecast failure times for optical fiber manufacturing equipment,increasing overall uptime by 3+ months. Multivariate Bayesian linear regression is used to gen-erate predictions and confidence intervals for machine health, which are visualized in PowerBIdashboards for plant personnel.Constrained Optimization for Ceramic Filters: Implemented and deployed Model PredictiveControl (MPC) algorithms for process control of extruder systems, resulting in reduced processdeviations by approximately 7%.

Conducted research focused on sparse optimal control methods for wide-area control in power systems. Frequency-based selective modal damping for power network oscillations allowed an improvement of stability by 5% over existing methods in the literature. Published papers in Control Engineering Practice and presented work at various technical conferences such as the American Controls Conference.

Development of a pre-built deep learning classification model AlexNet to separate defective parts. Deployment of a predictive process controller with Matlab optimization toolbox.

Assisted senior developers with PLC ladder logic, and emulating RSLogix PLC control blocks in the Matlab environment using Stateflow and Simulink.

Development of a distributed controls framework for a power system case study using off-the-shelf commercial devices including OrionLX from NovaTech, and Opal-RT real-time simulation platform.

Integrated a substation controller - OrionLX from NovaTech LLC, to the real-time computing hardware Opal-RT, through a TCP/IP connection through Modbus communication protocol.