尚未检测到更新。GitHub 发布版本后会显示在这里。
按最近已完成的 8 小时观测回合间的 GitHub 星标增量排序
Busabase is an open-source (MIT) database and workspace for AI agents — data, docs, skills, and apps in one place. Run it locally, self-host it inside your network, or use Busabase Cloud, and turn age
翻译准备中
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权，自行管理介绍，无需管理员批准。
登录并生成验证文件，在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件，即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录，将前往登录页面。登录后请回到此处，再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify，即可验证所有权并自行管理介绍。
一分钟完成验证正在加载完整 README。 以下为已保存的摘录。
Disaster intelligence for earlier action. A flood monitoring and response workspace for Ghana and Cameroon, built for the Pan-African AI Summit Hack-AI-Thon (Accra, 22–23 September 2026).
Avert turns scattered flood evidence — satellite-derived terrain, OpenStreetMap infrastructure, historical events, live rainfall and news reporting — into one operational picture: what happened, what's likely to happen next, and who to reach first.查看完整 README ↗Disaster intelligence for earlier action. A flood monitoring and response workspace for Ghana and Cameroon, built for the Pan-African AI Summit Hack-AI-Thon (Accra, 22–23 September 2026).
Avert turns scattered flood evidence — satellite-derived terrain, OpenStreetMap infrastructure, historical events, live rainfall and news reporting — into one operational picture: what happened, what's likely to happen next, and who to reach first.
Flood response in West and Central Africa is data-rich but evidence-scattered: terrain models, news photos, dam-release advisories and community reports live in different places and rarely get cross-referenced before a decision gets made. Avert's core bet is that a disciplined evidence model — every number tagged with where it came from and how confident we are in it — is worth more than a flashier dashboard that blurs the line between observed, modeled and reported.
| Screen | What it does |
|---|---|
| Monitor | Interactive map with historical flood reconstruction (day-by-day playback) and a forward-looking risk forecast, per community |
| Communities | Sortable, filterable table of every mapped settlement — exposure, road access, response priority — synced to the map |
| Evidence | Photographs, map layers and source reports behind each assessment, plus a live "Recent reports" feed of fresh incidents |
| Ask Avert | An AI assistant grounded only in the data on screen — cites its sources, explains scores, never invents numbers |
| Alerts | A three-step community-warning composer with a deterministic SMS delivery simulation (audience → message → review → results) |
| Account / Preferences | Demo identity, saved regions, map appearance, units, language, motion and notification preferences |
| Compact preview | A QR-linked, phone-sized version of the assessment view for field use |
real, curated, modeled or plannedEvery figure in Avert carries a provenance tag — OBSERVED, FORECAST, INFERRED, ASSESSED, REPORTED or UNVERIFIED — so an operator (or a judge) can always see whether a number came from a sensor, a model, or a news article. Response-priority and next-flood-risk scores are interpretable weighted indices with named, auditable factors — not a black-box model dressed up as one. The intelligence pipeline panel in Monitor walks through exactly how a score is built, end to end.
A contextual assistant scoped to exactly what's on screen: it receives a structured snapshot of the current region, event and selected community (never a full dataset dump), and is instructed to cite only sources it was actually given, flag missing inputs instead of guessing, and never invent evacuation orders, shelters or emergency numbers. It runs through a small server-side adapter compatible with OpenAI and Azure OpenAI — the API key never touches the browser. Without a key configured, Ask Avert shows an honest "unavailable" state instead of pretending to work.
The alert composer produces a fictional-recipient SMS delivery simulation — deterministic, seeded, with reconcilable delivery/failure counts and retry accounting. No real message is ever sent; every simulated message carries an unmistakable AVERT DEMO marker.
React 18 · TypeScript · Vite · MapLibre GL JS · Zustand · Turf.js · Tailwind CSS v4 · Motion
No backend database — regional/community data ships as static GeoJSON, app state lives in the browser, and the one server-side piece (Ask Avert) is a small framework-agnostic handler that runs identically under Vite's dev server and as a Vercel serverless function.
npm install
npm run dev
Open http://localhost:5173. Sign in with the demo workspace option on the login screen — no account required.
Copy .env.example to .env.local and set:
OPENAI_API_KEY=your-key
OPENAI_BASE_URL= # leave blank for api.openai.com, or set an Azure OpenAI v1 endpoint
OPENAI_MODEL=gpt-4o-mini # or your Azure deployment name
npm run build # typecheck + production build
npm run test # vitest — simulation engine + SMS segmentation invariants
npm run preview # serve the production build locally
Deploys to Vercel with zero configuration — Vite is auto-detected and api/ask-avert.ts is picked up automatically as a serverless function. Add the three OPENAI_* variables above under Project Settings → Environment Variables to enable Ask Avert in production; everything else works without them.
src/
components/ shell, map, inspector, chat, alerts UI
pages/ routed screens (Monitor lives in App.tsx; the rest are here)
data/ static geo/event/source registries
services/ exposure, forecast, weather and AI-context logic
domain/ pure alert-simulation and SMS-segmentation logic (tested)
stores/ alerts and preferences state (Zustand)
server/ shared Ask Avert handler (dev middleware + Vercel function both use this)
api/ Vercel serverless entrypoint
docs/ design system and screen specifications for this refactor
record/ provenance for the September 2026 evidence pack merged into the app
Built by Audran-wol for the Pan-African AI Summit Hack-AI-Thon, September 2026.