Tag: system design
-
We Should Plan to Fail
Photo by Zach Lezniewicz on Unsplash For weeks, the system behaved exactly as designed. Aria had matured from a directory of experiments into a stable local agent: custom tools, a conversational interface, a growing memory layer, scheduled journaling, and enough reliability to test new ideas without worrying about what might break. Luna, the quieter counterpart…
-
Building Luna: Part 1 – A Machine Learning (Through) Experience
Photo by Guzmán Barquín on Unsplash Readers following the Building Aria series probably expected Part 6 last weekend. I did too… until I made myself stop. Not from burnout or technical failure, but from a deeper question: what would actually be worth building next? So I imposed a two-week pause. No new features, no late-night…
-
Building Aria: Part 4 – The Alignment Mirror
Photo by Avery Evans on Unsplash Over the previous three entries, Project Aria has crossed three clear thresholds, though none of them felt clear while I was in the middle of them. In Part 1, she found her voice: more than a demo, she became a partner in conversation that carried beyond a single reply.…
-
Building Aria: Part 2 – From Voice to Agency
Picking Up the Thread In Part 1, Aria reached her first milestone: she was live in a practical sense. Running locally in Docker, serving models through Ollama, and speaking through Discord, she could already respond in conversation. Messages flowed into Postgres, the last ten entries recalled with each new exchange, giving her a baseline of…
Abstract Foundations 