Semantic map indexes 66,000 podcasts and 700,000 ideas
g_pal · reddit · 2026-08-04
A semantic map built from 66k podcast episodes and 700k extracted ideas
The author built a pipeline that transcribes podcast episodes, extracts discrete ideas, embeds and clusters them into topics, and plots them on a 2D semantic map.
- The map covers 66,000 episodes published over the last five months and 700,000+ discrete ideas.
- It lets users zoom into a topic, track discussion volume over time, follow key developments, and listen to the original clips.
- One of the hardest engineering problems was filtering out AI-generated podcasts; about 15% of financial podcasts in the pipeline contained synthetic content.
- Purely synthetic shows were relatively easy to remove, but mixed real + AI shows were harder.
- The project is available at sonicalpha.ai/atlas.
More from Apps
- JakeFromHeyGen made Lenny’s Product Pass launch video in 5 prompt iterations — lennysan · 2026-08-04
- Hamster adds global multi-chat and context graphs for product teams — EyalToledano · 2026-08-04
- Robot product builders should let users test on their own Hugging Face data — DominiqueCAPaul · 2026-08-04
- Every shows how voice plus AI agents can patch bugs and write work — every · 2026-08-04
- OpenAI says it rebuilt its voice stack end to end, from client to model — borowcy · 2026-08-04
- Microsoft Copilot’s business signup path takes users through multiple accounts and redirects — every · 2026-08-04