Lectrove turns hours of course and lecture video into time-aligned, checked data: a word-timed transcript, the full still of every slide and board page, notes, questions, decks and datasets ready for search, RAG and fine-tuning.
Speech to text with a second listen on unclear passages and consistent terminology.
Complete stills of each page, with verbatim screen text read by vision models.
Claude reviews each segment against the evidence, so errors are caught, not copied.
Chapters, notes, flashcards, slide decks and JSONL datasets, plus an API and MCP server for agents.
Claude is the quality core: it judges every segment, reconciles disagreements between models, writes notes and questions, and runs the agentic review loop after each batch. Cheaper local models do the bulk work; Claude makes the result reliable.
Bootstrapped, founded in 2026. The pipeline runs in production on our own GPU infrastructure and has mined full multi-lecture courses end to end.