The OCR & Speech Workspace is built on two Mistral AI capabilities: OCR that turns a scanned image or a 400-page textbook into structured, searchable markdown, and real-time speech-to-text with a post-recording accuracy refinement pass.
Page-batched, not page-by-page
OCR runs concurrently across page batches rather than one page at a time, with live per-page progress streamed over SSE. A book-length scan doesn't mean a coffee break — it means a progress bar that actually moves.
Chat that cites the page it read
Once a document is OCR'd, it becomes queryable — document-scoped RAG chat with citations that point back to the specific page a claim came from. The pattern is the same one from RAG.NextUpgrad: an answer without a citation is a claim you can't check.
Live transcription, refined after the fact
Speech-to-text streams live as you talk, then a second pass over the completed recording corrects errors the real-time pass couldn't catch — the same tradeoff every voice product makes between latency and accuracy, made explicit instead of hidden.
Repo and architecture: github.com/Pranjulrathour/OCR-STT-NEXTUPGRAD-mistral.ai-.
Pranjul Rathour · GenAI engineer, 3x hackathon winner, campus mentor. Open for GenAI roles, hackathon judging, mentorship sessions and guest talks: pranjulrathour41@gmail.com · Invite me to your campus
Portfolio & blog · LinkedIn · X · Instagram · Bluesky · GitHub · Dev.to
Top comments (0)