Bryan Etzine

ServiceNow | Santa Clara, CA | Python, PyTorch, & Docker

• Worked on the Core LLM team fine-tuning models with Fully Sharded Data Parallel on NVIDIA GPU cloud.
• Researched layer-wise quantization and dynamic pruning, shrinking model size up to 25% with similar accuracy.
• Utilized PyTorch Distributed Data Parallel to parallelize benchmarking in proprietary evaluation framework.

Machine Learning and Sensing Lab | Gainesville, FL | Python & PyTorch

• Labeled 2,000 fast Fourier transformed fish sound spectrograms, offering insights into coral reef aquatic life.
• Wrote KNN algorithm with Euclidean distance metric to classify spectrograms with 90% test-set accuracy.

Scale AI | Remote | C++

• Created prompt-answer pairs of programming tasks in C++ to fine-tune LLMs for optimal performance.
• Engaged in RLHF as a programming expert to align fine-tuned LLMs with human preferences.

ServiceNow | Santa Clara, CA | Java & JavaScript

• Resolved server timing issues by debugging race conditions across distributed nodes in a cluster.
• Patched a vulnerability in OAuth setup and reduced memory usage by properly unregistering profiles.
• Won 1st in hackathon for generative AI and autocomplete in server-side scripts, implemented in production.

ServiceNow | Santa Clara, CA | Java & JavaScript

• Implemented REST APIs for a HTTP client (95% unit test coverage) to manage multiple client connections.
• Improved streaming performance by 100% with lambda in-place stream modification approach.

Encoda | Tampa, FL | Python and SQL

• Created DevOps pipelines and automations to improve product deployment efficiency using Python, PostgreSQL, the Box API, and the Rundeck tool automating 3 hours of daily work manually updating the PostgreSQL database.
• Participated in an Agile team, attending daily standups, code reviews, assisting with sprint planning meetings, and learned how to realistically scope work for timely completion based on their expected difficulty.