Portrait of Justin Stamets

Hi! I'm Justin Stamets

I'm a full-stack engineer supporting health and wellness research, with interests spanning human performance, cognition, and computational modelling. In my professional work, I help research teams navigate IRB processes and data-sharing agreements, query operational and electronic health record data, and apply predictive modeling and causal inference to research questions. My broader interests include computational approaches to understanding affective, cognitive, and sensorimotor processes.


Please check out my Resume if you want to see my professional background, or view my projects involving cognition and machine learning: Affective Computing, Musical Computing, and Sensorimotor Computing, or simply hit the play button for a concise summary:

Pythia — a little desert jerboa in a gold emblem

Meet Pythia

The little-mouse icon on the opens Pythia, this site's resident assistant — a small language model that runs at the edge (a local Ollama behind the hardware bridge) or from an AWS-hosted endpoint. She was chosen as a tribute to the many small intelligences that are sacrificed for scientific discovery. She serves four primary functions — or to see her in action:

guide

Question answering about the pages and projects here — what the SEED model ablation found, how the guitar transcriber works, what the drone sim optimizes — from any page.

lab assistant

On the Affective Computing page she runs the SEED protocol end to end: checks your sensors, starts and paces the trials, collects self-reports, and archives the recording.

tutor

Practice another language with Pythia. Tutor Mode supports English, Spanish, and Mandarin, answering questions about the project side-by-side with English, providing pinyin for Mandarin, offering gentle corrections, and speaking with language-specific voices.

software maintainer

Reactive, personalized software requires constant iteration, revision, and maintenance. LLMs have accelerated many aspects of software engineering and design, and this site would have been considerably more difficult to build and maintain without them.

Much of the content and styling here emerged through repeated cycles of human and machine judgment, generation, and refinement. Perfection should not be expected, and neither should infallibility from Pythia. There is an ancient irony in consulting an Oracle: even prescient answers can mislead when the question, assumptions, or interpretation are wrong [Herodotus].

IncompleteStack Architecture — how this site runs: cloud and local, side by side A diagram with two parallel columns. From your browser the flow forks, and which fork it takes is a live choice rather than a fixed one. The cloud column shows a request flowing through an AWS load balancer and firewall to on-demand Fargate computers and a Bedrock language model. The local column shows the same flow through the hardware bridge at 127.0.0.1 to your own computer and a local Ollama model. Both columns run identical Docker containers, shown as matching chips feeding into two shared notes: same software packed once, and the lane it runs in being yours to pick at any moment. IncompleteStack Architecture Your browser your call, any time IN THE CLOUD scales with demand ON YOUR MACHINE private by design Front door checks and routes every visitor AWS ALB + WAF Hardware bridge a doorway only this computer sees 127.0.0.1 · LOCALHOST On-demand computers rented by the second, as needed AWS ECS FARGATE Your own computer the very same software, at home THE LAB LAPTOP sitepythiaspeech sitepythiaspeech identical Cloud mind Pythia's model, on a private line AMAZON BEDROCK Local mind Pythia's model, on this device OLLAMA More visitors? More copies appear — then vanish when the crowd leaves. Recordings stream straight to disk — nothing ever leaves the room. github.com/jstamet/local-lab ↗ Same software, packed once Docker seals each piece into a standardized container, so identical code runs on rented cloud machines or the lab laptop — the page can't tell which. Which lane runs it is your choice — every time Pythia's ⚙ picks her mind — hosted, or the bridge on your own machine — and the lab's “record to” picks where a session lands. Neither is fixed.