We process podcast transcripts and other datasets, use that to match over time based on similarity in meaning (or optionally exact and fuzzy matches if desired), then feed matching excerpts (match + surrounding context) into an LLM to score whether the speaker / writer agrees with {customizeable_poll_query_and_scoring}

:)

Note: similar to how DeepSeek started (or pandas or pick one of the many data science innovations), we actually are a quant investing software and data company. As a result of that, everything is time-series.