Notes
Observations from the work.
Research, engineering decisions, and lessons from building AI and infrastructure systems in production.
What More Thinking Actually Buys a Small Model
I tested whether small recurrent models can turn extra inference steps into deeper algorithmic reasoning. Extra steps do carry accurate prediction a few hops farther, but only after training across a range of depths—and my graph generator leaks structural cues, so I cannot yet tell whether that is the algorithm running or a shortcut.
What Code LLMs Actually Lack Is Navigation
I ran controlled experiments to figure out why code LLMs hallucinate cross-file references. The answer: they lack navigation, not comprehension. 14 tokens of symbol names gets 76% of the way. Full files get 92%. And you can inject the signal without any text at all.
Building a Browser-Based Video Editor with FFmpeg WASM
A browser-based video editor that runs entirely client-side using FFmpeg compiled to WebAssembly — no uploads, no servers, full privacy.
The Unexpected Benefits of Better Monitoring
Investing in Prometheus and Grafana did more than fix outages — it revealed performance bottlenecks, improved reliability, and changed how we think about building software.
Building Our First AI-Powered Feature
How we built and deployed our first AI feature — an auto-grading tool powered by a fine-tuned LLM — from model selection to production.
Why We Moved to Microservices
How we broke apart our monolith into microservices on Kubernetes — faster deployments, better reliability, and 40% lower infrastructure costs.