Now

Last updated April 24, 2026.

A /now page — what I'm focused on right now, not what I've ever done.

Building

• This site. Week 0. Pipeline-first: a posting cadence I can sustain matters more than any one post.
• A benchmark harness for consumer-GPU LLM evaluation. The first run is in post #3 when it lands.
• A hybrid Claude Code + Aider + Ollama workflow that automatically routes mechanical edits to a local model. Saves ~80% on API tokens for me.

Reading

• Anthropic's tool-use and prompt-caching docs — looking for places to cut latency on the planning side of the hybrid stack.
• The Aider source. Best way to understand the limits of small-model coding agents is to read the diff loop.

Not doing

• Twitter/X engagement farming. Time goes to writing, not threading.
• Side experiments outside the niche. One site, one beat, one year.

Stack right now

• Hardware: Desktop, RTX 3070 (8GB), WSL2 Ubuntu 24.04.
• Local model: qwen2.5-coder:7b via Ollama.
• Cloud model: Claude (the writing partner for this site).
• Site: Astro 5, GitHub Pages, zero hosting cost.

If something here changes and you'd like to know, RSS is the easiest way.