Auto-transcribe, summarize, and publish Taiwan's top finance podcasts — zero manual effort
Multiple daily podcasts covering Taiwan stocks — impossible to listen to everything
Key market analysis and stock picks buried in hours of audio content
No efficient way to scan key takeaways without listening to full episodes
PodSight automatically processes new episodes end-to-end, delivering readable summaries to your browser and Telegram
Taiwan's most popular stock market podcast. Deep analysis, market commentary.
Morning market briefing with macro analysis and economic outlook.
Afternoon market recap with sector rotation and stock picks.
6-step automated pipeline from RSS feed to your Telegram
Fetches each podcast's RSS feed and extracts episode metadata (title, date, audio URL)
Compares RSS episodes against existing summaries — only processes what's new
Downloads MP3 files for new episodes (audio is gitignored, re-downloaded as needed)
Powered by Groq's Whisper API — fast, accurate speech-to-text for Mandarin audio
Whisper large-v3 on Groq hardware — transcribes 1-hour episodes in seconds
Complete transcript saved as .txt — searchable, archivable, version-controlled in Git
Gemini processes the full transcript and generates a structured, readable summary
Generates a clean, searchable website with all summaries organized by podcast and episode. Auto-deploys on git push.
podsight.twNew episodes are pushed to the @podsight Telegram channel with a preview message and direct link to the full summary.
@podsightGitHub Actions runs the entire pipeline on a schedule — no human intervention needed
Speech-to-text with Whisper large-v3 on Groq's fast inference
Long-context AI summarization — handles full transcripts in one pass
Modular scripts for each step — easy to debug and extend
Static site hosting — auto-deploys on every git push
CI/CD scheduling — runs pipeline 2x daily automatically
Push notifications with formatted messages and direct links
A 3,300-line Python script generates 136+ static HTML pages — zero frameworks, zero build step
Single script outputs self-contained HTML with embedded CSS, vanilla JS, and Lucide icons via CDN. No compilation needed.
3 distinct color palettes injected via CSS custom properties. Each podcast gets its own visual identity automatically.
Handles TLDR, topics, stocks, humor, quotes, risks, and more — across 3 different AI summary output formats.
Click any stock tag to find all episodes mentioning it. Client-side regex filtering across 3 podcasts — no backend needed.
Compares feed against existing summaries, not audio — works across fresh CI environments
Waits 3 min after git push for Vercel to deploy, then verifies URL before sending
.telegram_published tracking file committed to Git — prevents re-sending old episodes
Handles EP#### (gooaye/zhaohua) and date-based (yutinghao) formats gracefully
Real production bugs from the first week of automation
Three ways to get your daily podcast summaries
Visit podsight.tw for all summaries, searchable by podcast and episode
Subscribe to @podsight for instant push notifications on new episodes
Full transcripts in Git — search, grep, or reference any episode's raw text