Back to projects

Skiply: Timestamp-Accurate AI Video Search

An AI-powered learning platform where a plain-English question returns ranked result cards that deep-link to the exact second a topic is taught inside a course video, playing on-site in the provider's own player, never on YouTube. Search runs as a three-phase server pipeline: an LLM writes CMS queries through the Sanity Context MCP, the server resolves a timestamp by matching keywords against video chapters then transcripts, and a second LLM streams back grounded, ranked cards that can never invent a course, lesson, or timestamp. Also includes Clerk auth with gated playback, watch-gated progress tracking, a "My Learning" page, and an offline yt-dlp pipeline that ingests transcripts and chapter markers into the CMS, all powered by the Vercel AI SDK and the Sanity Context MCP.

Skiply: Timestamp-Accurate AI Video Search

Key Features

Tech Stack

Next.js 16TypeScriptSanity CMSVercel AI SDKOpenAIModel Context ProtocolClerkPostHogTailwind CSSGROQ