Four students building a framework, cloud platform, and two research papers. GitHub Actions, pull request reviews, and issue tracking kept us aligned across all of it.
We needed a home for our trained models and datasets. HuggingFace Hub is where the ML community looks first, and their transformers library is the backbone of our training pipeline.
We needed serverless GPU access for RL training without managing infrastructure. Modal lets us spin up L40S and A10 GPUs with a Python decorator, no Dockerfiles or cluster configs required.