W
Willie singleton
About
Detail
Punjab, Pakistan
Résumé
Jobs
verified_user
0% verified
Senior Software EngineerArteraOct 2022 - Current (3 years 7 months)Designed preprocessing pipelines for multilingual text data using SentencePiece tokenization, language ID models, and quality filtering; stored aligned datasets in Bigtable for large-scale training. Trained Transformer models in TensorFlow on distributed TPU pods, applying mixed precision and gradient checkpointing to reduce runtime and cost. Fine-tuned with LoRA adapters and applied knowledge distillation to create smaller models optimized with TensorRT for low-latency inference. Deployed models through a FastAPI + gRPC backend on Kubernetes, with autoscaling, request batching, and versioned rollouts. Built an internal React + TypeScript evaluation tool for reviewers to score model outputs, annotate edge cases, and log results to PostgreSQ- Senior Software EngineerNikeSep 2015 - Oct 2022 (7 years 2 months)Designed and trained large-scale recommendation systems in PyTorch, leveraging user-item embeddings and sequence models to drive personalized feeds in the Nike shopping app and email campaigns. Built and maintained inference services with Flask and gRPC, deployed on AWS EKS with autoscaling and Redis caching to handle traffic spikes during sneaker launches. Developed demand forecasting models using LSTMs and XGBoost, trained on multi-year sales, marketing, and inventory data; orchestrated recurring training jobs with Airflow and stored results in Snowflake for downstream analytics. Improved product discovery by deploying BERT-based semantic search models, fine-tuned on Nike's product catalog and query logs, and integrated into the e-commerc
- Machine Learning EngineerKLAVIYOFeb 2013 - Sep 2015 (2 years 8 months)Worked on large-scale ranking models for the News Feed, training gradient boosted trees and deep learning models in Torch7 and early Caffe to improve content relevance and engagement. Built data preprocessing pipelines with Hive and Presto, generating billions of user-item interaction features daily and storing them in HDFS for model training. Deployed models into production using Klaviyo's internal prediction service APIs, adding versioning and shadow testing to validate new models before rollout. Collaborated with infrastructure engineers to optimize distributed training jobs on GPU clusters, tuning minibatch sizes and data sharding to reduce training runtime. Developed an internal React.js tool for data scientists and product managers to
- Data ScientistPwCAug 2011 - Feb 2013 (1 year 7 months)Developed predictive models for retail and banking clients using scikit-learn and R, focusing on churn prediction, fraud detection, and customer segmentation. Built ETL pipelines in SQL and Informatica to clean and aggregate transaction and CRM data for model training and reporting. Implemented classification and regression models (logistic regression, random forests, gradient boosting) and delivered insights through Tableau dashboards. Wrote Python scripts to automate feature engineering, model evaluation, and report generation, reducing manual analysis time for project teams. Collaborated with client engineering teams to deploy models as Java-based scoring services, integrating outputs into existing business applications. Assisted senior
Education
verified_user
0% verified
- M.Phil. (Master's) in Artificial IntelligenceUniversity of CambridgeMay 2009 - Jun 2011 (2 years 2 months)Research focus on machine learning and knowledge graphs
- B.A. in Computer ScienceUniversity of CambridgeAug 2005 - May 2009 (3 years 10 months)