Talk: Private Intelligence for Modern Practices
I gave this talk at Jersey Shore Tech and Design about building software that processes sensitive data without sending it to the cloud.
The presentation walks through a chiropractic notation system I’ve been building. Practitioners dictate exam findings, and the system extracts ICD-10 codes, CPT codes, and vertebral level references - all running on Apple Silicon using MLX.
The interesting part is the architecture. Three MLX models run in parallel and complete their analysis in under a second. The results feed into Apple Intelligence for clinical summaries, but everything stays on the device. Patient data never leaves.
The broader point: you don’t always need frontier models or cloud APIs. For many domain-specific problems, smaller models running locally are faster, cheaper, and keep data where it belongs.