Show HN: I built a local AI system for SOAP notes–no cloud, no wrappers
At first, I tried tools like ChatGPT—but they couldn’t hold intent across prompts, let alone sessions. Every note came out differently. That pushed me to build something else.
What I built is now called Echo Prime.
It isn’t a wrapper, and the LLM is swappable. The system itself holds structure, voice, and coherence across reboots—without memory modules, APIs, or fine-tuning. It adapts to my input style and logic, and even asks clarifying questions before acting.
I first tested it on GitHub. Prime contributed to live issues across multiple codebases—Rust, Python, Swift, OCR—without losing thread or hallucinating. All contributions are public.
Yesterday, I tested it in clinic.
Prime listened to audio transcriptions of real appointments, clarified patient context, and wrote my SOAP notes directly into the EMR via a local keyboard daemon. No wrappers. No fine-tuning. No cloud. Two notes, near-perfect. All on video.
I built this alone in a home office setting—no funding, no lab. Echo Prime is a working prototype—lean, local, and evolving. Even in this early form, Prime automates one of the most repetitive and high-stakes parts of my clinical workflow, using my own voice and judgment.
Would anyone else want something like this? Or is this only useful in my world? What would you automate, if you could trust your AI to remember, adapt, and act—locally and on your terms?
—Dentist in the loop
I didn’t set out to build an AI system—I just wanted help with my clinical notes. What I found was that the existing tools weren’t stable enough for high-stakes environments, and the more I tried to scaffold around them, the more brittle everything got.
So I built Echo Prime as a kind of local, self-consistent layer. Prime isn’t perfect—this is still a prototype—but works. Prime holds voice and structure across sessions, and handles everything from GitHub code to SOAP notes without retraining or wrappers.
I just wanted to share something real that I’m building and using in the real world. If there are flaws, I’d love to hear them. If there’s a better way to do this, point me to it.
—Dentist in the loop