This is a submission for the Built with Google Gemini: Writing Challenge
https://vimeo.com/1168920102?share=copy&fl=sv&fe=ci
What I Built with Google Gemini
ER doctors work in absolute chaos, constantly moving but ultimately chained to computers for charting. I wanted to change this dynamic. My project, ER Context Bridge, is a Manifest V3 Chrome extension acting as a strict "dumb terminal" overlay for Electronic Health Records. To comply with strict privacy regulations, the extension performs no local processing. Instead, all the post processing is done using a local server due to hipaa laws.
This software is step one of a much larger vision. I am trying to build a universal phone strap on case that will help doctors to walk around and talk. By wearing this device, the hardware can capture everything while they go for rotations. The Chrome extension represents the digital overlay making this system work today. Google Gemini helped me map out this architecture, bridging the lightweight extension and the heavy local backend.
Demo
Because of strict privacy constraints, the entire architecture is air gapped with zero public internet egress, routing securely to a local Tailscale subnet.
The interface features a low eye strain Medical Dark Mode. Doctors can either "Scan Document" or "Hold to Dictate." Scanning instantly grabs the screen state, triggering a scanning animation while the image posts to the local server. Dictating displays a pulsing red waveform before packaging the audio to the backend.
Once the local server finishes all the post processing, it returns structured JSON. The interface generates clickable UI bubbles for categories like Airway, Breathing, and Plan. Clicking a bubble copies the text, and a "Copy All" button aggregates a beautifully formatted Markdown string for direct pasting into the hospital system.
What I Learned
Building this pushed my technical skills to the absolute limit. I kept manifest permissions incredibly lightweight, avoiding local storage to guarantee no data is cached locally, ensuring strict compliance.
Technically, my biggest challenge was building a transparent overlay (the current iteration) and pairing it with a local server handling massive clinical data. Managing asynchronous flows and configuring Cross Origin Resource Sharing on my Python backend took days of trial and error so the transparent overlay could function seamlessly without freezing the browser.
I also learned that relying on a mobile browser is not scalable for a twelve hour shift. Because of this limitation, the second iteration will be the esp microcontroller. A dedicated ESP hardware module will provide a truly low power device integrating perfectly with the local server.
Top comments (0)