Oral-lens

Oral-lens is a Meta-powered evidence-review workspace for oral cancer researchers and student literature reviewers. Literature reviewers often need to determine whether a research statement is actually supported by the papers it cites. Conventional AI answers can sound confident while hiding missing details, weak retrieval, or unsupported conclusions. Oral-lens turns each research claim into an inspectable evidence trail. A reviewer can browse the existing research-paper corpus, search stored document content, preview exact passages, and manually select the papers that define the review scope. The application retrieves passages only from those selected papers and asks Meta Muse Spark 1.3 for a structured claim assessment. When the supplied evidence is incomplete, Meta can identify the missing information and propose a focused follow-up query. A bounded LangGraph workflow controls this process: Python enforces the selected-paper scope, limits the workflow to two searches, rejects repeated evidence, and mechanically validates cited passage IDs and quotations. If the evidence remains inadequate, Oral-lens reports insufficient evidence instead of inventing an answer. The React interface preserves browser-local review threads and visualizes each execution as a selectable tree containing scope classification, searches, assessment, cited passages, and source checks. Model assessment, mechanical validation, and human-review status remain separate. Oral-lens is a literature-research tool. It does not diagnose patients, evaluate oral lesions, recommend treatment, or provide medical advice. The earlier Oral Cancer X project supplied the inherited PostgreSQL/pgvector corpus, embeddings, and visualization foundation. During the hackathon, we built the Meta-powered review workflow, structured validation, bounded follow-up search, CLI/API integration, execution tree, evidence inspector, and new React research workspace.