About Me

Photo of Jillian outdoors in a forest. Jillian is sitting on a large, curved tree branch that rises several feet above the ground. They are facing slightly to the right with their body angled toward the camera. They are wearing a gray short‑sleeve shirt, tan hiking pants, black shoes, a black baseball cap, and a purple neck gaiter around their neck. They have a green backpack with visible shoulder straps. Their hands rest on the branch for balance. The forest around them is filled with tall pine trees, and sunlight filters through the branches, creating bright patches on the ground. The setting appears calm and natural, with no other people or objects visible. The photo captures a moment of rest during a hike, showing Jillian in a stable seated position on the branch, surrounded by trees and natural light.
Long‑form DeafBlind alt text describing Jillian seated on a curved tree branch in a sunlit pine forest.

I'm Jillian! I'm a Deaf, Neurodivergent, Disabled, non‑traditional coder based in Colorado. No, I don't live in the pipelines. Why do you ask?? I build local‑first, community‑first tools on Raspberry Pi, AMD ROCm, and open-source stacks. My work centers accessibility, autonomy, and community ownership.

My public GitHub work lives here:

Spoonie Helper — Current Status

Spoonie Helper 🥄 is my community‑first assistive‑technology price comparison tool. It prioritizes Deaf‑owned, Disabled‑owned, Black‑owned, Indigenous‑owned, POC‑owned vendors, etc., before major retailers by design, and without the ability to be turned off.

The backend evolved from my earlier project AccessiFind: a vendor‑finder, classifier, and ranking engine built on a Raspberry Pi agent, PostgreSQL, and a two‑model LLM stack (Qwen2.5‑3B + Phi‑3.5‑mini). It now powers Spoonie Helper’s natural‑language AT search, community‑first scoring, and HTML reporting.

Current status:
The ingestion pipeline, vendor directory, and ranking engine are complete.
The static frontend and Fuse.js search UI are in progress.
The Pi agent is running locally and handling background tasks.

Local Compute & Model Training

I maintain a stable home training environment built around a Raspberry Pi 3B+ and an AMD ROCm Ubuntu PC. The Pi hosts TinyLlama as a lightweight local model server and background agent. On the GPU side, I’m actively training Qwen 2.5–3B and a fresh TinyLlama variant using QLoRA for fast iteration and low‑power fine‑tuning.

This setup powers my local‑first workflow: rapid prototyping on the Pi, nightly training on the ROCm machine, and fully offline inference for accessibility‑focused tools.

Reach Me

Email: jg@jg18.dev