Empowered Intelligence

Work

This page covers shipped systems, delivered websites, and the one you are reading.

Docatar

An AI-powered patient intake platform. Docatar talks with the patient before the appointment and hands the clinician a structured summary, so the visit starts at the problem instead of the paperwork. Application and website designed and built in house.

A language model in front of patients is the hardest version of the integration problem: the conversation has to adapt to the person while never exceeding what an intake is allowed to be. A bounded conversation, structured output the clinician can act on, and one team owning every surface from the model to the public site.

Status
In provider beta, docatar.com
Role
Product, engineering, and site. In house.
Notable
Conversational intake, structured clinician summaries, no forms or app download

Sanctuary Alert

A personal-safety iOS application built and shipped to the App Store. Encrypted local storage, cryptographic integrity verification for captured media, and a bilingual interface. Designed, built, and published in house.

Personal-safety software must work the first time, under stress, for someone who has never rehearsed. So data stays on the device, encrypted at rest, and captured media carries cryptographic integrity verification, so it can be trusted later by people whose job is to doubt it. Shipped in house: design, engineering, App Store review, and every failure report since. There is no version of that work where a mistake belongs to someone else.

Platform
iOS, shipped to the App Store
Role
Design, engineering, release. In house.
Notable
Encrypted local storage, cryptographic media integrity, bilingual interface

AI validation harness

A validation layer for a production AI system that constrains model output, verifies it against specification, and surfaces failures before they reach a user.

A system like this does not fail by crashing. It fails by answering fluently and wrongly. The harness constrains what the model may produce, verifies each output against the written specification, and logs every rejection with its reason. An AI system is not done when it works in a demo. It is done when someone can state what it will not do.

Role
Design and engineering of the validation layer
Client
Confidential

Web development

Client and business websites, delivered and live. Each was built end to end: design, markup, search, and hosting and DNS provisioned in the client's own accounts.

Product sites for Docatar and for Sanctuary Alert, the latter bilingual with its public privacy documentation. Timberwolf Spinners, a Pennsylvania maker of handmade fishing spinners: full catalog, species guidance, and an ordering flow built around how the business actually sells. Empowered Recordings, the founder's audio studio since 2012: custom HTML, technical SEO, and a rates page that shows prices without making you ask. Marshall's Madness, a small-batch Maryland hot sauce maker: a storefront with cart and inventory, the seed-to-sauce story, and a pairing guide.

Live
docatar.com, empoweredsanctuary.com, timberwolfspinners.com, empoweredrecordings.com, marshallsmadnesshotsauce.com
Role
Design, build, SEO, and hosting and DNS setup in the client's accounts

This site

The page you are reading is an exhibit too. It renders a real-time field behind every page and still holds itself to a stated budget: 2.5 seconds to first paint on a 4G connection, 16.7 milliseconds per frame.

The readout on the home page is not a graphic. It is the site measuring itself on your device, against those budgets, while you decide whether to hire us. If WebGL is blocked, every page still reads and navigates. This is what rigor looks like when it is published rather than promised.

Standard
WCAG 2.1 AA, Section 508, verified by measurement
Degradation
WebGL to 2D canvas to plain text, all navigable
Telemetry
Measured on your device, live, on the home page

In development

QuestScreen, a Meta Quest 3 application that streams multiple computers into a headset as virtual screens: a Unity VR client, cross-platform desktop agents, and a dual-transport pipeline over WebRTC and USB H.264 with a GPU decode path built for latency.

Yours could be next.

Start a conversation