Datasets
Global media transcripts dataset (earnings, news, podcasts)
Spoken finance is where positioning is explained and hedged. A CFO's tone on the third analyst question, a central banker's aside at a conference, a commodity trader on a regional radio segment — none of it appears in a filing, and almost none of it is searchable in a form an agent can cite.
Hindcast is building a transcript layer across earnings calls, world news broadcasts, radio and finance podcasts: speaker-attributed, time-coded to the second, and chunked so retrieval returns a passage rather than an hour. Every extract carries the program, the speaker and the broadcast timestamp, which is what makes a quote checkable instead of merely plausible.
Language coverage is part of the point. A Portuguese-language interview about a harvest and an English earnings call about the same commodity belong in the same index, joined by the entity graph rather than by hope.
hindcast — session
previewhindcast> find where management discussed pricing on the last earnings call
→ tool: transcripts_search { entity:"XYZ", type:"earnings_call", query:"pricing" }
✓ matching passages with speaker and timestamp
source: call transcript, time-coded to the second
What's covered
- Earnings calls with prepared remarks and Q&A separated by speaker
- World news and radio segments with program and broadcast timestamp
- Finance podcasts indexed at passage level rather than episode level
- Speaker attribution and role where the source identifies them
- Multi-language coverage joined through the shared entity graph
- Passage-level citations: program, speaker, timestamp on every extract
Join the waitlist — this page's dataset is already selected.
Questions
Are transcripts machine-generated or official?
Where an official transcript exists it is preferred and marked as such; otherwise the transcript is generated and labeled, so an agent can weight a quote by how it was produced instead of treating all text as equally reliable.
Can I search across languages?
That is the intent — entities and topics are resolved across languages, so a query about a company returns relevant passages whether they were spoken in English or Portuguese.
Related
Join the waitlist. Vote the roadmap. First in, first served.