Machine Learning Engineer2018 - PRESENTZulily
Technologies: Java, Spark, H20, Kubernetes
- Adapted an API for deploying a scalable, cloud-based machine learning model (Go/Kubernetes/Docker).
- Developed a driver for communicating with said API (Apache Airflow).
- Wrote a back-end process for pulling data into an in-memory cache (Java/SQL).
- Developed EDA and validation metrics (Java/H20).
Machine Learning Engineer2018 - 2018Boldiq
Technologies: C, C++, DLIB
- Developed an AI-based optimization engine employing deep reinforcement learning for learning strategies for optimized resource-constrained scheduling.
- Maintained and debugged ‘Solver’, company’s proprietary real-time optimization engine (for private aviation scheduling).
Senior Data Scientist2016 - 2017Cisco Systems
Technologies: Python, NumPy, Pandas, SciPy, Matplotlib, SciKit-Learn, CVXOPT, Keras, TensorFlow, Splunk
- Oversaw AI for optimizing network monitoring. Developed a machine learning pipeline allowing for analysis of Cisco’s unstructured data (through Splunk’s Rest API) using ensemble techniques from the SciKit-Learn library. Initiative resulted in improved event correlations on Cisco CMS’s network management platform.
- Extended the initiative to perform network incident forecasting using deep learning techniques on a customized architecture (NLP, semantic analysis via CNNs) using a TensorFlow backend and Keras (high-level API).
- Architected of “Splunk to Excel," an automated reporting mechanism.
Intern/Research Assistant2015 - 2016Ecole Polytechnique Federale de Lausanne
Technologies: Cal, Java, Spring, lp_solve, MySQL
- Developed embedded DB and SDC-constrained scheduling software in Java for High-Level Synthesis and data-flow programming applications (with applications to embedded systems). Accepted into the doctoral program of the EPFL.
Intern/Research Assistant2014 - 2015Swisscom
Technologies: Matlab, Optimization toolkit, Signal processing toolkit, Transmission line theory, Oscilloscope
- Developed insertion loss and cross-talk cable models for 4th-generation DSL standard, G.fast. The use of vectoring to achieve higher data rates requires an accurate understanding of how the cable manipulates intended signaling. Models were used in the Broadband Forum for standardization purposes.