Blog

Ramblings

Your inference provider sucks (at caching)

The same agent session cost $0.13 on one DeepSeek V4 Flash provider and $2.30 on another. Prompt cache hit rate, not list price, explains the 17.7× gap.

Sep 2, 2026 · 5 min read

The Area Over the Curve

Why the cheapest LLM is rarely the cheapest result, and how August routes each agent task to just enough model intelligence.

Jul 13, 2026 · 4 min read

Everything is a Workflow

How workflows became August's abstraction for reliable legal AI agents, growing from saved prompts into a small programming language.

Jul 9, 2026 · 4 min read

HOPS

A discrete-event simulator for pipeline-parallel LLM training on heterogeneous GPUs, validated within 2% of real hardware.

Apr 20, 2026 · 7 min read

Frame Classification in Congressional Speeches with NLP

Using NLP to classify rhetorical framing in U.S. Congressional speeches, our University of Michigan CSE 595 final project.

Feb 4, 2026 · 16 min read

An Honest Review of Every ULCS Class I Have Taken at Michigan

Honest reviews and ratings for every upper-level CS (ULCS) class at the University of Michigan.

Feb 3, 2026 · 3 min read