Experience

Ivani LLC

• United States
• Software Development
• 1 - 100 Employee

• Senior Machine Learning Engineer

  • Apr 2021 - Present

  Residential Technology - Team LeadCompleted a 6-month cycle of a Beta test, algorithm development, and evaluation. Guided the team towards key metrics and determined priorities to balance timelines and deliverables. Communicated weekly with external stakeholders to present progress and distill the business goals that drove the technology cycle.

• Machine Learning Engineer

  • Jan 2019 - Apr 2021

  User Experience - Team LeadFocused on user-oriented design to create a variety of meaningful data views, including: RESTful APIs, Dashboards, Indoor mapping.Platform TechnologyExpanded and maintained a data server pipeline capable of processing data from 10,000+ connected edge devices. Emphasized automation and robustness through a test-driven development workflow to reduce engineering time and improve system reliability.

• Data Engineer

  • Aug 2018 - Jan 2019

  State-space modelingLeveraged time-series information to improve the robustness of machine learning models to classify occupancy state.Testing and validationDesigned barrage of tests to identify edge cases and failure points to determine robustness and recovery of machine learning models.Data collection and testingAutomated data collection and pre-processing pipeline to validate machine learning models, testing edge cases and failure points. As the scale of the project grew, worked with new technologies to stream high-throughput data.



University of Pittsburgh

• United States
• Higher Education
• 700 & Above Employee

• Graduate Student Researcher

  • Sep 2010 - Jul 2018

  Data scientist skilled at developing and evaluating predictive models of time-series and behavioral data. Predicted behavior from brain data using computational models that leveraged multivariate regression and advanced linear algebra techniques. Evaluated models with statistical metrics, bootstrapped confidence intervals, and model selection. Experience collecting and storing data from complex sources, while also developing analysis tools aimed at scaleable and flexible exploration of the data.Leveraged multivariate regression models to establish a logical flow of hypothesis testing to classify information content. Cross-validated prediction results efficiently replicated key metrics in time-series data that defined future experimental behavior.Manipulated and explored 100-dimensional data set to visualize simple patterns using dimensionality reduction. Also used linear algebra to find structure within the null dimensions of a linear model, re-defining the model parameters and capturing behaviorally relevant variability that was previously ignored.Found the optimal form of an autoregressive model of time-series data using the Laplace transform and ordinary differential equations. Tested model performance to verify prediction smoothness and to determine the best number of parameters. The model made it possible to achieve results with current equipment, avoiding costly purchases and project delays.Described complex human behavior in terms of a physical dynamical system with high explanatory power and few parameters. Tested hypotheses about the damping and stability of the system and found that, at the limits of response times, the system was under-damped and oscillated, leading to insights about how humans learn to build predictive internal models.

• Teaching Assistant

  • Jan 2011 - May 2012

  Voted by the students as the teaching assistant of the year for the Bioengineering Department.



Washington University in St. Louis

• United States
• Higher Education
• 700 & Above Employee

• Research Assistant

  • Jun 2009 - May 2010

  I began my research experience as a volunteer with Jack Engsberg, Ph.D. in the Human Performance Laboratory (Program in Occupational Therapy at Washington University in St. Louis School of Medicine). I calibrated a SmartWheel device used to study repetitive motion injuries in wheelchair users. To do this, I designed and carried out a unique calibration process using known forces and the voltage output signals produced by the force transducer. In addition, I collaborated with occupational therapists and students to track and plot human body motion. These data were used to assess hippotherapy research and to evaluate the gait pattern of stroke patients testing a mechanized boot for therapy and assisted walking. Later I worked as a research assistant with Daniel Moran, Ph.D. (Department of Biomedical Engineering at Washington University in St. Louis). I combined computational modeling, musculoskeletal dynamics, and computer science in a framework designed to explore functional electrical stimulation (FES) as a clinical intervention that would enable paralyzed individuals to perform activities of daily living. Toward this goal, I qualitatively improved the anatomical accuracy of 39 musculotendon paths in a 3 dimensional, 7 degree-of-freedom musculoskeletal arm model. I then used the model to simulate activities of daily living and estimate muscle activity. In the model, 39 musculotendon units were used to actuate the arm; however, stimulating 39 muscles with FES is a daunting task for even the most advanced systems. My goal was to find the smallest set of muscles that was able to perform the desired movement. I collaborated with a computer science graduate student to employ techniques designed to speed up the simulation’s run-time and together we were able to identify a set of 24 muscles, reduced from the original 39, that would lift a cup and bring it to the mouth for a drink.