Too many unread books — auto-extract structured knowledge graphs from texts

Too many knowledge workers—estimated 200 million globally—struggle to convert the books they buy into usable, queryable knowledge: longform texts are dense, fragmented across topics, and time-consuming to synthesize, which leaves insights locked behind unread pages and passive highlights. The result is wasted learning time, duplicated research, and decision-makers who rely on summaries that often omit provenance or granular claims. The product would automatically extract structured knowledge graphs from books and longform texts, surfacing entities, claims, evidence, and citations, and exporting interoperable nodes and links into PKM tools like Obsidian, Notion, and Roam. Architecturally it would combine embeddings for dense retrieval, LLMs for information extraction, domain-specific parsers for tables/figures, and a human-in-the-loop verification layer to deliver high-precision facts, provenance, and cross-book linking at a usable cost model (targeting a $300 ACV per user or team). This is a timely opportunity: advancements in LLMs and embeddings materially lower inference cost and raise extraction quality, and a $60.0B addressable market (200M knowledge workers × $300 ACV) with a Market Score of 95/100 and Revenue Potential at 82/100 signals commercial viability. Competition is medium—there are summary and note-taking tools—but standing out requires measurable accuracy, rigorous provenance, standards for export, and enterprise-grade integrations; challenges include extraction fidelity across genres, copyright/compliance, and earning user trust, so initial focus should be high-value verticals and tight PKM interoperability rather than a broad consumer play.