Professional Experience
Senior Software Engineer (MLOps + Backend)
Ford Pro, Ford Motor Company (Jun 2021 – Present)
- Evangelized containerization; built MLOps team from scratch; influenced talent mobility, architecture, data policy
- Architected and deployed microservices on cloud platforms (GCP), built CI/CD pipelines and Infrastructure as Code (Terraform, Tekton, Helm) using VertexAI, PubSub, Cloud Run, Cloud Functions saving $200k/year
- Built fullstack web applications (Spring Boot, FastAPI, NextJS, React) in production using TDD
- Proactively broke down business-critical legacy monolith services worth over $100M into microservices handling over 3-5M req/month using GraphQL and ElasticSearch
- Developed Machine Learning systems using Vertex AI, Kubeflow, and Airflow, saving over $27Million/year
- Built & piloted fullstack Retrieval Augmented Generation (RAG) application using Large Language Model (LLM)
- Identified and mitigated security vulnerabilities in cloud-native applications using SonarQube, 42Crunch
- Wrote FastAPI auth middleware library using OAuth2, JWT, and integrated with enterprise IAM systems
Software Developer (Data + Algorithms)
Ford Motor Company (May 2021 – Jun 2021)
- Designed and implemented data pipelines for real-time data processing using Apache Kafka, Spark, and Hadoop
- Developed algorithms for data analysis and machine learning model training
- Collaborated with cross-functional teams to integrate data solutions into existing systems
Software Engineer (Mobility Research + ML Systems)
Metropia Inc. (Sep 2015 – Mar 2018)
- Helped get Google and Apple app store rating over 4-stars by finding and fixing backend and smartphone app UI
- Built and tested SOTA Texas border crossing time prediction ML feature for smartphone app for a $408k project
- R&D’d advanced routing and navigation algorithms using Python+Java, created services using Neo4J, Kafka
Graduate Researcher (ML Research + Algorithms)
University of Connecticut (Aug 2013 – Aug 2015)
- Developed a software prototype similar to Google Maps' routing engine using crowdsourced GPS data
- Created a prototype algorithmic framework for estimating route travel time
- Simulated hypothetical crowdsourcing scenarios in Python to evaluate model efficacy
- Implemented Hidden Markov Model-based GPS map-matching algorithm with 95% accuracy using a novel Confusion Matrix approach
- Achieved over 90% accuracy in route travel time estimation using Regularized Least Squares Regression
Details
Implemented algorithms to infer transit vehicle routes from GPS readings and detect deviations from predefined routes, providing accurate route detection every minute. Developed a Hidden Markov Model (HMM) based map-matching algorithm using the Viterbi algorithm for filtering noisy, low-frequency GPS data. All code was implemented in Python using ArcPy, open-source GIS libraries, NumPy, SciPy, and Pandas. The main objective was to predict Estimated Time of Arrival (ETA) and analyze crowd motivation.
Simulated a crowdsourcing environment to predict ETA as a Trajectory Regression problem, solved with Kernel Ridge Regression. Utilized open-source GIS libraries such as GDAL/OGR, Shapely, Fiona, Pyproj, NetworkX, PyKML, ArcMap, Network Analyst, and Matplotlib for data processing and visualization. The project is hosted on GitHub.