Thoughts on AI engineering, automation, and building products from the ground up.
Working with LLMs taught me that the best tools don't just generate output — they reason through problems. That shift changed how I approach every project I take on.
2026Running models on your own hardware isn't just about privacy. It's about understanding what you're building. You learn more when nothing is abstracted away from you.
2026A single prompt rarely solves real problems. Breaking tasks into stages — discovery, research, generation — creates systems that actually produce consistent, reliable results.
2026Most chatbots answer from training data. Adding a web search layer transforms them into something genuinely useful — answers grounded in current, verifiable information.
2026Building Qlaude and Curio as a solo developer meant owning every layer — auth, billing, infrastructure. It's harder, but you understand your product completely.
2026RAGE doesn't just generate posts. It discovers, researches, writes, and designs. Automation works best when each step has clear purpose and validation built in.
2026