Sterling Jones
Machine Learning Engineer
Los Angeles,CA
Summary
Machine Learning Engineer focused on real world trust and safety problems. Possesses experience in applied deep learning, predictive algorithms, and quantitative analysis. Various experience in both small stage startups and larger corporations. Proven success working within fast-paced, collaborative settings. Experience with all aspects of ML pipelines from training to deployment.
Overview
4
4
years of professional experience
Work History
Trust and Safety, Machine Learning Engineer
Tinder
3 2022 - Current
- Trained, deployed, and scaled state of the art deep learning models for detecting "bad actor" images that receives over 3 million requests per day with a p99 latency of 900 ms.
- Further expanded NLP capabilities using LLM's and country specific tokenization. Additionally, worked on rare token detection for identifying scam/spam within text.
- Implemented a new pub/sub kafka infrastructure for image data. This reduced latency and solved traffic spikes caused by outages and flink backpressure.
- Built a custom training pipeline on AWS to solve GPU data starvation. By building a custom data loader pipeline to stream images continuously, I was able to keep GPU utilization at 100% and reduce training time by a factor of 3x.
Embedded Machine Learning Engineer
Amberbox Inc
07.2021 - 03.2022
- Prototyped various on edge machine learning models and quickly determined application viability. Specifically, utilized CNN models for anomaly detection within audio files.
- Built an end to end data pipeline for preproccesing, augmenting, and labeling millions of audio files.
- Researched best practices and applied synthetic data techniques to balance training datasets. As a company focused on gunshot detection, positive samples were scarce compared to negatives. By utilizing Synthetic Minority Oversampling Technique and similar techniques, I was able to generate enough data for training deep neural networks.
Machine Learning Engineer (Part-time)
Everipedia
05.2020 - 06.2021
- Achieved classification accuracy of 95% by implementing CNNs, LSTMs, and various models for EEG brainwave data.
- Created a low latency Brain Computer Interface. This read EEG data from a wearable device, preprocessed it, and fed it into custom algorithms for interacting with a display in real time.
- Increased model accuracy on average of 15% by automating feature selection using a custom Genetic Algorithm. Brainwave data can posses non-linear relationships, making PCA and similar linear feature selection algorithms ineffective.
Education
Bachelor of Science - Cognitive Science With Computing Specialization
University of California - Los Angeles
Aug 2017 - 06.2021
Skills
Python and SQL
Machine Learning and Deep Learning
Computer Vision
Natural Language Processing
Time series analysis and signals processing
ML Infrastructure and Deployment
Pyspark and distributed computing
Jenkins, Scaffold, Airflow
AWS, Docker, Kubernetes
Fine-tuning and transfer learning
Timeline
Embedded Machine Learning Engineer
Amberbox Inc
07.2021 - 03.2022
Machine Learning Engineer (Part-time)
Everipedia
05.2020 - 06.2021
Trust and Safety, Machine Learning Engineer
Tinder
3 2022 - Current
Bachelor of Science - Cognitive Science With Computing Specialization
University of California - Los Angeles
Aug 2017 - 06.2021