Thoughts and Explorations
Writing about software engineering, AI, product thinking, and the craft of building things.
My perception of AI changed from "it can do anything" to "it can only do one thing at a time"
How splitting one overworked model into two, one that talks and one that thinks, took my tool call failures from 15% to near zero.
Inspiration for a system architecture came from a long drive
How a graph-based navigation engine lets my voice AI process an entire sentence worth of information in one turn instead of making callers repeat themselves.
My AI lied to me. Here's how I made it stop.
How I built three invisible systems that catch failures, predict responses, and stop my voice AI from confidently lying to callers.
My AI Could Talk. It Couldn't Listen. That Was the Whole Problem.
How I built three layers of active listening that make callers feel heard, from back-channel "mm-hmm"s to instant fillers to responses that bypass the model entirely.
The Architecture of Conversation States for LLMs
How a state machine architecture fixes the biggest problem in voice AI — and why giving a small AI less to think about makes it dramatically smarter.
Hello World, My First Blog Post
A quick intro to this blog and what I plan to write about. Software engineering, AI, and building things that matter.