How I Think About AI
"I don't build wrappers around APIs and call them AI products. I build cognitive architectures that solve genuinely hard problems — memory systems that forget, research engines that think, infrastructure that actually ships. The future of AI isn't just smarter models. It's smarter memory, smarter tooling, and builders who understand both the neuroscience and the engineering trade-offs."
cognitive infra in production. fewer wrappers, more substrate. the page below dives into one build at a time.
A neuroscience-inspired memory engine that forgets, strengthens, corrects, and consolidates — just like the human brain.
Three memory types modelled on the established neuroscience taxonomy — episodic, semantic, procedural — each with biologically plausible decay rates and reinforcement dynamics. Ebbinghaus decay curves, memory reconsolidation, episodic-to-semantic consolidation, and trust-weighted retrieval, all behind a JSON-only HTTP surface designed for agent tool-use.