SEMINAR: Synchronization Engine for Mathematical Illustration, Narration, and Reasoning

Translating dense Mathematical literature into an accessible format that an audience can easily follow is a task routinely performed by human Mathematicians. Our goal is to computationally replicate this pedagogical experience for any Mathematical literature, as reading Mathematical papers is cognitively demanding. Achieving this replication presents several significant challenges. First, current Text-to-Speech (TTS) models fail to parse Mathematics semantically, defaulting to rigid, symbol-by-symbol recitation. However, even perfect audio translation is insufficient, as comprehending Mathematics requires referencing prior derivations and formulas; therefore, synchronous visual writing is essential. Generating this synchronized writing for Mathematics is particularly difficult due to the complexity of typographic symbols and the frequent inclusion of geometric diagrams. Furthermore, orchestrating these multimodal elements into a perfectly synchronized presentation introduces substantial temporal scaffolding hurdles. To address these challenges, we present SEMINAR (Synchronization Engine for Mathematical Illustration, Narration, and Reasoning), a browser-native system that transforms arbitrary Mathematical papers into synchronized audio-visual seminars. Finally, we propose, but have not yet implemented, a verification stage that autoformalizes the source in Lean 4 and checks that each claim in the generated seminar is entailed by it, blocking unsupported assertions and flagging the rest for review.