Brendan Rappazzo
Bio
Brendan Hogan is a machine learning research scientist in Morgan Stanley's ML Research group, where he works on LLM fine-tuning, reinforcement learning, and agentic workflows for frontier models. He holds a PhD in Computer Science from Cornell, where he worked with Carla Gomes on Computational Sustainability.
Sessions (2)
Brendan Rappazzo (Brendan R. Hogan) is an AI/ML research scientist at Morgan Stanley (PhD, Cornell) focused on LLMs and agentic systems, with recent work including AlphaLab, an autonomous multi-agent research harness that generated GPU kernels averaging 4.4x faster than torch.compile and achieved 22% lower LLM pretraining validation loss, making his session a compelling look at frontier agentic AI applied to rigorous quantitative research. His open-source projects like hermitclaw and loophole show a pattern of building self-directed AI systems, suggesting a talk rich with practical insights on autonomous agents.
GitHub
@brendanhogan- brendanhogan/loophole★ 392Adversarial moral-legal code system that stress-tests ethical principles and system prompts by finding loopholes using competing AI agents with a judge agent.
- brendanhogan/hermitclaw★ 327A tiny autonomous AI creature that lives in a folder, continuously researching, writing reports and notes, and developing ideas without user prompts.
- brendanhogan/DeepSeekRL-Extended★ 253Recreates DeepSeek R1's RL approach (GRPO) from scratch at small scale, training Qwen 1.5B on grade school math.
- brendanhogan/2025_advent_of_small_ml★ 22A 25-day series of self-contained small ML experiments covering steering vectors, prompt optimization, unsupervised training, and more.
- brendanhogan/completion_tree_view★ 15Lightweight Python library for visualizing how language models generate text at the token level via interactive HTML or PDF tree visualizations.
Recent writing (2)
Public activity researched automatically · as of Jun 2026