Literature Map Builder

Researchers spend the first week of any project reading in a random order, which hides the shape of the field. Embedding-based clustering finds the shape before you commit to a reading order, so the gap you end up chasing is one of the field's actual gaps — not just a gap in your own reading.

1. Embed each paper's abstract + methods section with a strong retrieval model. 2. HDBSCAN over the embeddings to find natural clusters (no k to pre-specify). 3. For each cluster, sample 3-5 papers into a context window and ask the LLM for a methodology+findings synthesis with a critical-gap bullet. 4. Second pass: compare the per-cluster gaps against all papers' conclusions to rule out gaps that are actually addressed elsewhere. 5. Emit a markdown report with per-cluster sections and a consolidated 'open questions' list at the bottom.