Transcript Seeker is a project I did for Meeting BaaS, and I think it was my first one there. The idea was simple: one app where you can manage and look through all your Meeting BaaS meetings, read the transcripts and ask questions about them.
What it does
Meeting Baas sends bots into calls, and Gladia or AssemblyAI transcribe uploads. You bring your own API keys, and OpenAI answers your questions. Click a word in the transcript and it jumps to that moment.




How it went
The app already existed when I picked it up in October 2024. It was a React client talking to an Express API with its own Postgres database, plus a few bits stashed in IndexedDB. I moved the data into the browser with PGLite, then split the project into a Turborepo monorepo with the web app, the proxy, a small API and a docs site built with Fumadocs. Calendars came next, so you could see your upcoming meetings, and the last stretch was mostly fighting deployment configs. Right after this I built the Speaking Meeting Bot.
How it works
The fun part is the architecture. The core app is a client-side SPA with no backend of its own. Everything lives in PGLite, which is Postgres running right in your browser, with Drizzle on top. Your keys, meetings, notes and chats all stay on your machine. A tiny Nitro proxy passes requests through to the Meeting BaaS API. The one thing that does need a server is calendar sync, so a small API with Better Auth sign-in keeps calendar data in a SQLite database on Turso. Recordings play in Vidstack, and the transcript scrolls along with the video. The UI is built with shadcn/ui.
