Prepare podcast transcripts for retrieval with Grok Bot and Palantir Foundry
At a glance
Hans C. Nelson used Grok Bot to build a Foundry pipeline that transcribes YouTube podcasts, identifies speakers, and embeds meaningful transcript segments. This workflow is listed under transcription, data-pipelines, rag.
Nelson built the pipeline in a free developer account on Palantir Foundry with Grok Bot's help. It transcribes podcasts from YouTube, labels speakers, divides the transcript at changes of thought, and creates embeddings for a planned hybrid retrieval system.
How it went
Nelson reports better transcription quality than YouTube's native transcripts, without publishing a benchmark in the post.
Worth knowing
The hybrid retrieval component was still unfinished when he posted; the completed work described is transcript preparation and embedding.
Save ideas in this browser. Upvotes count across the board.
A builder pattern for selling Grok Bot work: give each client a studio+client@mail.grokbot.com tag and a routine that only accepts requests from approved contacts.
With Grok Bot’s own inbox live, a GTM builder sketched an always-on pipeline: Firecrawl monitors prospects, then the bot drafts client outreach for human send.
A builder shipped a Karpathy-inspired shared-spine memory for Grok Bot—one wiki, permanent bidirectional links, short/medium/long-term as tags (never copies), with a nightly tidy job.
After a Grok Bot cut one user’s Comcast bill by $43/mo, a coffee-shop tester gave Grok Bot, ChatGPT Dots, and Meta Muse the same retention/negotiation job to compare how agents handle real bills.
