Sebastian Sosa

- Designed a scalable ML pipeline using PyTorch, Ray, WandB, sklearn, and GCP, automating end-to-end model development for NNs, XGBoost, and Sklearn models.
- Engineered production implementations of interpretability algorithms for automated knowledge discovery workflows.
- Developed and open-sourced an LLM attribution library
- Established ML engineering standards (testing, code reviews, CI/CD)

- Developed a demand forecasting and price optimization system using XGBoost, increasing sales velocity by 4% and GMV by $9M.
- Implemented DevOps practices with GitHub Actions for CI/CD and leveraged GCP services for infrastructure management.
- Fine-tuned a transformer language model for topic classification and sentiment analysis, achieving 91% mean F1 score.

Created an automated recruiting Web SMS Messaging Platform using React, Node.js, and PostgreSQL. This platform streamlined the recruitment process by enabling efficient communication.

- Developed a 3D medical training application for Microsoft HoloLens using Unity3D.
- Implemented real-time mannequin position tracking and holographic procedure guidance, and developed a 360° video player app for Android & iOS.

Published a peer-reviewed paper on differential privacy in ML, An Assessment of the Application of Private Aggregation of Ensemble Models to Sensible Data as part of the graduation thesis.

Awarded a grant by Open Philanthropy for AI safety upskilling and independent research. Completed the Alignment Research Engineer Accelerator (ARENA) coursework. Topics included PyTorch, Transformers, Mechanistic Interpretability, Reinforcement Learning, and RLHF.