Selected software projects

Selected software projects spanning AI, health, and systems engineering.

NinetyBytes is a compact portfolio of private projects, prototypes, and technical demos. The work here ranges from on-device glucose prediction and AI-driven tooling to low-level systems software and local inference infrastructure.

Projects

Selected Projects

Each section is short on purpose. The goal is to give enough context for a resume reference without turning this into a full showcase site.

Retired macOS app

GlucoGram

Dexcom Share, macOS menu bar utility

A retired macOS menu bar utility formerly distributed through the App Store for viewing live Dexcom glucose data without opening a separate app.

  • Pulls glucose data from Dexcom Share and shows current value and trend at a glance.
  • Built from a personal need to reduce context switching while working.
  • No longer maintained and retired from the App Store in 2024; included here as a concise project reference.

Private iOS/watchOS AI app

Core ML and AI glucose prediction app

Core ML, AI, glucose prediction, pattern recognition, exercise impact

A private iPhone and Apple Watch app focused on glucose prediction, therapy context, and forecast-driven decision support built around on-device ML.

  • Centers on on-device glucose forecasting with Core ML, personalization, retrieval, and forecast-confidence work.
  • Explores exercise-impact previews and what-if workflows that show how planned activity may affect glucose before a workout starts.
  • Includes pattern-recognition and insight work over historical glucose and therapy data, alongside phone, widget, Live Activity, and Apple Watch surfaces.
  • Includes insights and trend surfaces for forecast accuracy, recurring time-of-day patterns, exercise impact, and post-bolus response built from merged CGM and therapy history.

AI + systems project

Custom x86-64 operating system with local AI runtime

C, x86-64 assembly (arm64 port in progress), Qwen, GGUF, local inference

A long-running from-scratch operating system spanning bootloader, kernel, scheduler, virtual memory, VFS, PE-style program loading, and a growing native userland, with local AI integrated directly into the platform.

  • Boot path from loader to kernel bring-up and user-mode runtime.
  • Integrated local Qwen-based inference and agent tooling using GGUF models, native runtime components, and queue-driven workflows.
  • NT-style section mapping, demand paging, and file-backed memory to support heavier native workloads.
  • Desktop shell, native apps, and iterative systems debugging.

Supporting systems work

Networking, desktop UI, and native tooling

IPv4, ICMP, UDP, TCP, TLS, window manager, native apps

Supporting platform work built alongside the operating system: networking, TLS, windowing, task inspection, and first-party native apps.

  • e1000-based IPv4 networking with ICMP, UDP, and TCP support.
  • TLS client tooling for secure outbound requests.
  • Window manager, Finder, and Task Manager style native desktop apps.

About

About NinetyBytes

NinetyBytes is the name I originally intended to use for an independent software company. It never became a formal company, but the name remains the umbrella I use for personal work across AI, health software, and systems engineering.

This site is intentionally selective. It captures a small set of projects clearly enough to show architecture, direction, and implementation depth without turning the page into a full case-study archive.

Contact

Contact

For follow-up, additional project context, or technical discussion, email is the simplest way to reach me.