FirstMinute The first minute decides the stroke. We make it count. A paramedic describes the patient in plain speech. FirstMinute scores the RACE stroke scale item by item with Jev (TypeSafe System One), asks back only what it is unsure about, checks live hospital status with Browserbase + Stagehand, and after the medic confirms opens a CODE STROKE iMessage group with the stroke team via Photon, with an SBAR handoff written on GMI Cloud. Pulsara and JoinTriage put the paper stroke form on a phone. FirstMinute removes the form. Jev scores. Code decides. The medic confirms. Synthetic data only; decision support, not a medical device. Run it Requires Node 22 (nvm use 22) and Python 3.10+. # 1. Backend (FastAPI) — :8000 cd backend && python3 -m venv venv && ./venv/bin/pip install -r requirements.txt cp .env.example .env # add TYPESAFE_API_KEY, GMI_*, BROWSERBASE_* when you have them ./venv/bin/uvicorn main:app --port 8000 # 2. Photon bridge (iMessage) — :8787 cd messaging && npm install && cp .env.example .env # SPECTRUM_PROJECT_ID / SPECTRUM_PROJECT_SECRET npm run build && npm start # 3. Frontend — :5173 cd frontend && npm install && npm run dev Open http://localhost:5173 (landing) and http://localhost:5173/console (product). The regional hospital status board lives at http://localhost:8000/portal. Every integration runs live when its key is set and simulated otherwise. The console shows which mode each one is in. Demo script (3 minutes) Landing (20 s): the problem in numbers. Console → Play "Clear LVO": the transcript streams in, RACE rows fill live, each Jev decision shows its latency, and the total reads 7/9 · LVO suspected. Play "Missing gaze": the gaze row goes dashed and FirstMinute asks "Are the eyes deviated?". Answer it and the row turns confident. Routing: St. Luke's is excluded because its angio suite is occupied (read from the status board), so the route goes to Mercy General. Open /portal, flip Mercy's Neuro IR off, and watch the route change. Confirm & pre-alert: the CODE STROKE group chat appears with the SBAR. With Photon live, it lands on the stroke team's real phones. Real phone: text the project's iMessage number as the medic. The same flow runs over iMessage, and the console mirrors it. Repo Path What backend/ FastAPI: Jev, routing rules, hospital status scraper + mock portal, GMI SBAR, Photon flow, SSE frontend/ React + Vite + TS + Tailwind + shadcn: landing page and live console messaging/ Photon Spectrum bridge (iMessage) video/ Remotion pitch film → video/out/firstminute-pitch.mp4 docs/SPEC.md Product + API contract + design brief Why Jev Jev returns typed answers with calibrated probabilities in about 100–400 ms. Clinical scales are typed forms, so each RACE item maps onto a Jev score question. All items plus "was this mentioned?" checks go in one parallel request. Low confidence triggers a follow-up question instead of a guess. Date math, glucose thresholds and routing stay in deterministic code, where they belong. https://youtu.be/EFZEMWtdc5g -- demo link
Built at JEVATHON (w/ The AI Collective)
FirstMinute
FirstMinute The first minute decides the stroke. We make it count. A paramedic describes the patient in plain speech. FirstMinute scores the RACE stroke scale item by item with Jev (TypeSafe System One), asks back only what it is unsure about, checks live hospital status with Browserbase + Stagehand, and after the medic confirms opens a CODE STROKE iMessage group with the stroke team via Photon, with an SBAR handoff written on GMI Cloud. Pulsara and JoinTriage put the paper stroke form on a phone. FirstMinute removes the form. Jev scores. Code decides. The medic confirms. Synthetic data only; decision support, not a medical device. Run it Requires Node 22 (nvm use 22) and Python 3.10+. # 1. Backend (FastAPI) — :8000 cd backend && python3 -m venv venv && ./venv/bin/pip install -r requirements.txt cp .env.example .env # add TYPESAFE_API_KEY, GMI_*, BROWSERBASE_* when you have them ./venv/bin/uvicorn main:app --port 8000 # 2. Photon bridge (iMessage) — :8787 cd messaging && npm install && cp .env.example .env # SPECTRUM_PROJECT_ID / SPECTRUM_PROJECT_SECRET npm run build && npm start # 3. Frontend — :5173 cd frontend && npm install && npm run dev Open http://localhost:5173 (landing) and http://localhost:5173/console (product). The regional hospital status board lives at http://localhost:8000/portal. Every integration runs live when its key is set and simulated otherwise. The console shows which mode each one is in. Demo script (3 minutes) Landing (20 s): the problem in numbers. Console → Play "Clear LVO": the transcript streams in, RACE rows fill live, each Jev decision shows its latency, and the total reads 7/9 · LVO suspected. Play "Missing gaze": the gaze row goes dashed and FirstMinute asks "Are the eyes deviated?". Answer it and the row turns confident. Routing: St. Luke's is excluded because its angio suite is occupied (read from the status board), so the route goes to Mercy General. Open /portal, flip Mercy's Neuro IR off, and watch the route change. Confirm & pre-alert: the CODE STROKE group chat appears with the SBAR. With Photon live, it lands on the stroke team's real phones. Real phone: text the project's iMessage number as the medic. The same flow runs over iMessage, and the console mirrors it. Repo Path What backend/ FastAPI: Jev, routing rules, hospital status scraper + mock portal, GMI SBAR, Photon flow, SSE frontend/ React + Vite + TS + Tailwind + shadcn: landing page and live console messaging/ Photon Spectrum bridge (iMessage) video/ Remotion pitch film → video/out/firstminute-pitch.mp4 docs/SPEC.md Product + API contract + design brief Why Jev Jev returns typed answers with calibrated probabilities in about 100–400 ms. Clinical scales are typed forms, so each RACE item maps onto a Jev score question. All items plus "was this mentioned?" checks go in one parallel request. Low confidence triggers a follow-up question instead of a guess. Date math, glucose thresholds and routing stay in deterministic code, where they belong. https://youtu.be/EFZEMWtdc5g -- demo link
Keep exploring what builders shipped.
Gustave
Gustave, the voice shopping assistance
Online shopping means typing, filtering, scrolling and comparing across tabs. And most "shopping assistants" just swap the search bar for a chat box: type, wait, see three products, type again, wait again. In a store, you just talk, and someone starts pulling things off the rack while you're still talking. Every reaction you give sharpens what they bring next. Meet Gustave: a voice shopping assistant that recreates that natural, iterative experience.
HackerSquad project
UnSlop
Slop isn't just AI, it can also be human. Unslop will input a website and cut out all the ineffective fluff/slop with context decisions and browserbase lookup centered around jev decision making.
KYC-Sentinel
KYC Sentinel
KYC Sentinel: adverse-media screening for AML/KYC analysts. Jev judges, code decides, the LLM only writes. Compliance analysts waste hours clearing irrelevant news hits: same-name strangers, victims, passing mentions. KYC Sentinel pulls news for every client in a portfolio. Jev (TypeSafe AI) judges each client–article pair: same person? the perpetrator? what risk? how severe? A pure, unit-tested router then sorts every client into CLEAR / REVIEW / FLAGGED / NO_COVERAGE. • Fails closed: any uncertainty or model error means REVIEW, never CLEAR. • Autonomy Dial: a compliance manager drags strictness and the whole portfolio re-routes instantly from stored Jev probabilities, with zero new model calls. • Thresholds live in code, not prompts. The LLM only writes analyst memos, on demand, for flagged/review clients, citing sources. It never decides. • Full audit log of every decision, plus a live receipts meter. Results on a 26-client portfolio: 37 Jev calls, 0 errors, 0.9s, $0.0011, about 145× cheaper than the same job on a frontier LLM (est. $0.16). Famous-name collisions (Michael Jordan, Michael Cohen) are cleared as "different person" even when the other person's news is damning. Victims and identity-theft targets are cleared as "not the subject". Vague common-name mentions go to REVIEW. Documented fraud cases and a Tulsa accountant who shares Taylor Swift's name are surfaced, while the pop star's news is ignored. Stack: Jev (TypeSafe AI), FastAPI, GDELT, Gemini for memos, vanilla JS. 77 tests. Data note: GDELT rate-limited our network, so several clients use clearly labelled synthetic articles (domain synthetic-demo.local). Everything downstream runs identically.