Andre Kiselev
Web 2.0 & Web 3.0 Full-Stack Product Builder
New to Contra
Andre is building their profile!
A 30-second launch video for Moonshot Legends, the paid research pass on Polysnipes.com. Built as a Python code - a deterministic renderer that outputs the finished 1080p file, with a sound design pipeline wired to the same source of truth. One command rebuilds the whole thing.
The result is a marketing asset that lives in the repo next to the product: versionable, diffable, reproducible by anyone who clones it.
1
53
Someone is choking, bleeding, or not breathing. There is no signal, no training, and about four seconds of usable attention. Every first-aid app I could find assumes the opposite: that you have a connection, that you can read a paragraph, and that you will trust it without ever showing where its instructions came from.
AidEmber is the version I wanted to exist. You describe what you see in plain words. It routes you to a pre-authored, source-cited protocol and walks you through it one decision per screen — four words, a metronome, a single action. Everything runs on the device.
The rule the whole thing is built on: the model picks WHICH protocol applies, and never writes the instruction. Every word on every screen is copied verbatim from a published resuscitation guideline, and a drawer on each step links back to the exact section it came from.
That rule earned itself during this build. Ten minutes in, the design model quietly rewrote a compression depth in its own words. It was not even wrong — but it was generated, not transcribed, and on a screen telling someone how to keep another person alive, that difference is the entire product. I caught it, restored the exact text, and put the constraint into the file's design guidelines so it could not happen again.
Nine screens: the triage home, the one clarifying question the app is allowed to ask, the call for help, the breathing check, compressions at 110 per minute, the source drawer, the infant choking protocol, and the hand-over. Same design system throughout.
Built in Flowstep, driven from my Claude Cowork with Opus 5 over MCP. The protocol engine, the format, the safety tests and every citation are open source.
Guidance only — not a medical device, and not reviewed by a medical board yet. That is the next step, and it is the honest state of the project.
Prototype: https://app.flowstep.ai/file?activeFileId=0bea1c04-ae74-4cd2-9897-834fc13405c7
(https://app.flowstep.ai/file?activeFileId=0bea1c04-ae74-4cd2-9897-834fc13405c7)LinkedIn Post: https://www.linkedin.com/posts/debythm_flowstepchallenge-activity-7489134115426820096-G-KS?
Repo: https://github.com/horn111/aidember
1
77