James Braza
Member Of Technical Staff - AI Research at Edison Scientific
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolJames Braza is an AI research engineer with a decade of hands-on experience building scientific agents and post-training LLMs, currently on the learning and platform teams at Edison Scientific after a spinout from FutureHouse. He brings a rare blend of ML research and production-grade engineering, with NeurIPS and ICLR–ranked work on scientific reasoning and self-improving foundation models alongside practical systems for retrieval-augmented recommendation engines. Previously he led instrument and control software at Synthego and built automation/test systems at SpaceX, often writing concurrent Python systems that interface with hardware and real-time sensors. An active open-source contributor, he’s improved robustness in projects like promptfoo (LLM prompt testing) and enhanced RL training in Hugging Face’s trl, reflecting a focus on reliability and reproducible ML workflows. Based in San Francisco, he favors rapid iteration, open collaboration, and engineering rigor—plus an offbeat Github bio that reminds you he’s human and enjoys spaghetti.
10 years of coding experience
8 years of employment as a software developer
B.S. Mechanical Engineering Economics Minor, B.S. Mechanical Engineering Economics Minor at University of PittsburghArtificial Intelligence Graduate Certificate, Artificial Intelligence Graduate Certificate at Stanford University
Github Skills (22)
transformers10
apim10
pytorch10
python10
evaluation10
machine-learning10
reinforcement-learning10
llm10
prompt-engineering10
eval10
api10
openai10
ollama10
modbus10
debug9
Programming languages (22)
Github contributions (5)
huggingface/trl
Apr 2024 - Mar 2025
Train transformer language models with reinforcement learning.
Role in this project:
ML Engineer
ML EngineerContributions:6 reviews, 6 PRs, 15 comments in 11 months
Contributions summary:James's contributions primarily focus on enhancing the training process within the `trl` library, specifically related to reinforcement learning with transformer models. The user implemented modifications to the training configuration, including passing custom BOS/EOS tokens, and addressed issues that arose when using certain features, such as `accelerate` and `deepspeed`. These changes also include type hinting adjustments and improvements to loss calculations, indicating a focus on improving performance, fixing bugs, and increasing flexibility.
nlptransformer-architecturelanguage-modelfinetuninglanguage-modeling
promptfoo/promptfoo
Sep 2023 - Aug 2024
Test your prompts, agents, and RAGs. Red teaming, pentesting, and vulnerability scanning for LLMs. Compare performance of GPT, Claude, Gemini, Llama, and more. Simple declarative configs with command line and CI/CD integration.
Role in this project:Backend Engineer
Backend EngineerContributions:6 reviews, 8 PRs, 87 comments in 10 months
Contributions summary:James contributed to the project by addressing failures in OpenAI, LocalAI, and Ollama embeddings traversal. They improved error handling, including the logging of API responses. Additionally, the user refactored code to align with environment variables, improving overall code maintainability. These efforts focused on improving the robustness of the promptfoo project, and the interaction with LLM API providers.
cici-cdcicdevaluationevaluation-framework
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.