DEMO GALLERYPREVIEW
Three small products, each rendered straight from the AiPod graph. No transcripts wrangled, no scraping, no model at query time — every number, quote, and timestamp below is real corpus data with the receipt attached, and each page rebuilds itself as new episodes land.
393 dated, falsifiable calls across 83 episodes — who said it, the deadline, and the verbatim receipt. Filterable live; the raw material for an accountability scoreboard.
OpenAI according to 26 episodes across 6 shows: 132 claims from 29 voices — including OpenAI’s own CFO — compiled onto one chronological page by the entity graph.
The podcast week, compiled by a script: a lead story, first mentions, and 20 new predictions from 2 episodes — regenerated on every ingest, no editor involved.
The hard part — transcribe, diarize, attribute, compile — happened at ingest. What's left for you is a template.
One endpoint returns structure, not audio: every claim attributed and timestamped, every quote verbatim, every prediction dated. covered: false when the corpus doesn't know — so your product never bluffs either.
The only code you write is a render loop. Each page in this gallery is one or two API calls and a template — no database, no pipeline, no prompt, nothing to moderate.
Weekly ingest moves every show's knowledge cutoff forward and these pages rebuild themselves. The graph compounds — so does whatever you build on it.
# 01 — the ledger curl 'https://aipod-two.vercel.app/api/episode/<id>' | jq '.predictions' # 02 — the dossier curl 'https://aipod-two.vercel.app/api/entity/OpenAI' # 03 — the briefing curl 'https://aipod-two.vercel.app/api/episodes'
MCP and plain JSON, no API key, $0 per query, cited to the second. Point your agent at it — or your weekend project.