Machine Learning Engineer - Production ML Systems

Theo Walker

Lilburn, Georgia • ************@*******.*** • 971-***-**** • www.linkedin.com/in/theo-walker- 62b89891

SUMMARY

Machine Learning Engineer with 15+ years building production systems, most recently at Roadie where Python-backed models cut routing errors across a network handling roughly 10,000 daily delivery requests. Prior work at Pindrop shipped fraud-detection pipelines on audio and metadata for over a billion phone calls, and earlier research at UNC Chapel Hill produced Chembench, an ML application for drug-activity prediction still used by 8,000+ researchers worldwide.

SKILLS

Languages & Frameworks: Python, SQL, Java, C++

ML & Data: scikit-learn, TensorFlow, PyTorch, pandas, NumPy, Spark, Airflow Cloud & Infrastructure: AWS (EC2, S3, EMR, Athena, DynamoDB, Batch, ECR), Docker, Terraform

Testing: Cypress, RestSharp

Practices & Collaboration: Agile/Scrum, Jira, Confluence, code review, mentoring, CI/CD, observability (Prometheus, Grafana, Cloud Logging) EXPERIENCE

Machine Learning Engineer Roadie GA, US 12/2022 – Present Shipped route-optimization models in Python that reduced late deliveries by roughly 18% across the carrier network while keeping driver-app latency under target thresholds.

Owned end-to-end ML pipelines from data ingestion through serving, covering feature engineering, model training, and Dockerized prediction endpoints with Prometheus and Grafana monitoring.

Partnered with ops and product to define model performance metrics, then added A/B testing hooks that let the team validate routing changes before full rollout. Mentored two junior engineers on Python tooling, code-review practices, and observability setup, cutting onboarding time for new ML features. Rebuilt the data-access layer for the ML platform, replacing ad-hoc scripts with parameterized SQL queries that cut analyst wait time by roughly 30%. Following an outage during peak volume, added automated alerting on pipeline error rates and latency so the team could catch regressions before they reached production. Senior Software Engineer Pindrop Atlanta, GA 11/2016 – 12/2021 Developed ML pipelines on audio and metadata that detect fraud across 1 billion+ phone calls, with Dockerized prediction microservices monitored for accuracy, cost, and errors.

Owned fraud-detection pipelines processing regulated financial and voice-biometric data subject to PCI-DSS and SOC 2 audit trails, implementing access controls and logging that passed annual compliance reviews.

Created Pindrop Trace, a graph-based fraud-detection system that uses machine learning on call relationships and earned a patent for the team. Built cloud infrastructure and data-access tools for the data science team, including IAM roles, S3 data pipelines, and Airflow DAGs that reduced manual ETL work. Improved pipeline reliability by adding Spark-based validation checks and alert routing through Jenkins pipelines, cutting incident mean-time-to-detect by around 25%.

Research Software Engineer Max Planck Florida Institute for Neuroscience Jupiter, FL 6/2011 – 10/2016

Built an interactive cell segmentation tool (polar transform, dynamic programming) in Python that remains state of the art and is used by thousands of researchers worldwide.

Developed an iterative 3D multimodal image registration and stitching pipeline using FFT and minimum spanning tree algorithms, supporting high-throughput neuroscience imaging workflows.

Designed and constructed an animal behavior training system with custom sensors, data acquisition, and real-time feedback loops for experimental neuroscience studies. Research Software Engineer University of North Carolina at Chapel Hill Chapel Hill, NC 07/2008 – 05/2011

Developed Chembench, a machine learning web application for predicting drug activity and toxicity, now used by 8,000+ researchers worldwide to identify medical compounds and potential toxins.

Implemented scalable data pipelines handling molecular descriptors and model outputs, enabling researchers to run large-scale virtual compound screening without manual processing.

EDUCATION

University of North Carolina at Chapel Hill 09/2006 – 05/2011 Master's Degree

Rensselaer Polytechnic Institute 09/2002 – 05/2006 Bachelor's Degree, Computer Science, Bioinformatics and Molecular Biology

Contact this candidate