Senior Software Engineer2017 - PRESENTConnected Signals
Technologies: Python, Pandas
- Provided prediction models for bus ETAs at traffic lights for signal priority requests.
- Created a prediction model for response delay for drivers at a traffic light turning from red to green.
- Created prediction models for traffic lights from vehicular probe data (GPS traces).
Contractor: Machine Learning Researcher2017 - PRESENTEduworks Corporation
Technologies: Python, Java
- Created a tractable algorithm to infer skill mastery in complex competency frameworks from potentially conflicting sources of information.
- Performed experiments with an algorithm on simulated data to validate the correctness and significance of improvement over baseline.
- Wrote a white paper on inference in complex competency frameworks.
Software Developer and Machine Learning Expert2016 - 2017Eduworks Corporation
Technologies: Java, git, Maven
- Created a fully automated multiple-choice quiz generation system, which takes in a text document and identifies keywords and generates questions.
- Developed a system to perform inference in complex competency frameworks, parsing (potentially conflicting) evidence from various sources.
- Built a machine learning system to detect voice phishing attacks from audio signals, by detecting various types of tampering (splicing, synthetic voices, replay attacks).
- Developed several ideas for potential project proposals, involving machine learning, natural language processing, and automatic decision making.
Postdoctoral Researcher2012 - 2015Oregon State University
- Created various algorithms for learning in an interactive loop between (human) subject matter expert and AI system.
- Designed various algorithms for imitation learning (a machine learning system that takes in expert solutions and tries to learn the underlying policy).
- Developed a system to perform path planning for multiple agents for a complex search and coverage problem.
- Created an algorithm to perform inference in MOOC frameworks by predicting readiness-to-learn, using Bayesian techniques.
- Designed algorithms for multi-task classification, by exploiting task descriptions and balancing specification and generalization.
Postdoctoral Researcher2011 - 2012University of Waterloo
- Developed algorithms to perform inference on a discrete output variable with a continuous input space.
- Built a system to perform inference and planning in a complex system to assist in post-stroke rehabilitation.