From meeting noise to structured knowledge: The hidden engineering challenges behind Sorinai

If the task of natural language processing is to extract order from the chaos of language, then meeting minutes are the ultimate sample of that chaos. Multiple speakers taking turns, topic jumps, unfinished sentences, implicit references—these features make semantic understanding after meeting transcription almost equivalent to identifying the source of every raindrop in a downpour. Sorinai positions itself as an "interactive AI notepad." This name itself implies a promise: it doesn't just record, but also understands, and allows users to interact with the recorded content. From a product logic perspective, this looks like trying to reconstruct a downpour into a river distribution map.