lifesage

LifeSage was a venture to build an AI life-coaching product, a chat app where a "Sage" persona held long-running coaching conversations with a user. It reached alpha-1 - deployed at lifesage.ai on Render, with analytics groundwork in place - before being paused. The Rails 7 + Hotwire stack carried disciplined fundamentals: ViewComponent, RSpec, Devise + Auth0, Tailwind/Haml templates, and a Person-centric domain where users and sages were both Person records linked through conversations and messages.

The more interesting part was the memory architecture. A family of Derivative* models captured asynchronously-computed summaries of conversations and person profiles, and every LLM call logged its token usage. That progressive-compaction approach - compressing long histories into retrievable artifacts - turned out to prefigure ideas I came back to years later in callstack. Alpha-2 plans (vector-DB long-term memory, dynamic summary retrieval) were scoped but not shipped.