Skip to Main Content

Soumya Das

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Soumya Das — Demo video

KinBeat finds exposed keys, tokens and passwords before hackers do. Secrets leak every day in the places people actually work: a chat message, a PDF, a config file, a public web page, a GitHub repository, a pull request, a Google Drive file, a Slack channel. In 2025 alone, over 28 million new secrets leaked on public GitHub. KinBeat is a free website with no sign-up and nothing stored. You paste a chat, upload a file or screenshot, paste a website link, paste a GitHub repository link, scan your pull request history, or connect Google Drive or Slack. KinBeat finds every secret, including ones hidden in base64, hex, number words or split over two messages, and shows the file, page or line, with the value masked. The hard part is not finding candidates but deciding what matters. Meta's Muse Spark judges every finding as real, possible or only an example, with a one-line reason, so you fix the real ones first. A validation layer checks every model answer before it is shown. KinBeat AI, in the same chat, answers questions about your own findings and about what a key is and how to rotate it. The whole pipeline is scored by an eval suite of over 230,000 cases; the latest live run scored 98.9 percent. Built solo in ten days. Public repositories you do not own only show counts, so nobody can use KinBeat to raid other people's code; a read-only token unlocks your own.

Team