Safer multi-tenancy with Postgres's row-level security

Pushing tenant scoping from application code into the database.

Multi-Turn RL for Code Debugging

Training a 7B model to debug code in a custom DSL environment using GRPO. Comparing prompting, supervised fine-tuning, and reinforcement learning across three categories of bugs.

Visualizing Adam's adaptive learning rates

An interactive visualization showing how Adam adapts its step sizes in response to different gradient patterns.