Writing

Leaderboard gaps in Kaggle reasoning challenges are mostly a compute budget problem, not a skill gap. Three tiers, one framework.

On a multi-week agent build, chat context resets every session. Root-level docs like ARCHITECTURE.md and COMMONMISTAKES.md are how the model kept state.

How I turned an Obsidian vault into a persistent memory and identity layer for every Cursor session, across every project, without bloating context.

Why building games with AI in 2026 is an exercise in system design and "AI ergonomics," not just syntax.

How I used AI to go from a generic white portfolio to something that actually reflects what I do, focusing on product decisions rather than implementation.

Can AI models know what they don't know, and why that gap matters for AGI.

How hybrid search and structured knowledge turn a scattered Obsidian vault into a high-fidelity long-term memory for AI agents.

How I built a real-time dashboard to track LLM API spend before it tracks me.

The agentic workflow that took a tool from idea to terminal in one afternoon.

A portfolio site, cost dashboards, automated reports, and a personal AI, in seven days.

How I set up OpenClaw on Docker and configured a personal AI that actually knows my context.