Learn the market in 5 minutes a day.
The best finance shows are long, fast and full of jargon. Stocksplain listens to them for you and turns each one into a short, plain-English study page, with the market data beside it.
Read today’s digest →Why it exists
Too much to follow
Hours of podcasts and hundreds of headlines every day. Nobody learning to invest has time for all of it.
Too much jargon
Basis points, convertible notes, P/E ratios… Every term is explained in plain English, with an example, and saved to a growing glossary.
Too much noise
No hype channels or stock tips. Only hosts and newsrooms with real credentials, with opinions clearly separated from facts.
How it works
- 1
Find
Every morning we check 15 hand-picked podcasts and 27 newsroom feeds for anything new.
- 2
Listen
Each new episode is transcribed word for word by an AI speech model (Google's Gemini in the cloud, or open-source Whisper as an offline backup).
- 3
Understand
An AI model reads the transcript like a teacher and writes the key points, the important numbers, the lesson and the jargon to know. It is told to use only what the source said, to keep forecasts and opinions separate, and to flag low-value episodes instead of summarizing them.
- 4
Connect
Headlines from across the newsrooms are grouped into the day’s hottest stories, and prices, charts and headlines are added for the companies involved.
- 5
Publish
Everything becomes one daily page, archived forever, with a searchable glossary of every term so far.
What to keep in mind
- It’s AI-written. Summaries are checked for accuracy, but mistakes can happen. Every story links to the original, so check it before relying on a number.
- It’s for learning, not advice. Nothing here is a recommendation to buy or sell anything.
- Credit goes to the creators. We publish short summaries in our own words, never transcripts or full articles, and always link back. If you enjoy a show, listen to it or subscribe. Stocksplain is not affiliated with any show, newsroom or company mentioned.
- Market data comes from Yahoo Finance, refreshes every 15 minutes while the US market is open, and may be delayed.
- Creators: if you’d like your show removed, open a request and we’ll take it down promptly.
How it's built
Stocksplain is a full-stack product run on a few free cloud services. Here's what's going on behind the screens.
🧠 Serverless AI, $0 to run
Every morning a cloud job wakes up, listens to the new episodes with Google's Gemini models and writes the day's content: no laptop, no servers to run. If a model is busy or out of free quota it falls back to the next one, and the same pipeline can run fully offline on a Mac with open-source models (Whisper + Qwen 14B). Because the AI works once per day rather than once per visitor, cost doesn't grow with the number of users.
🔎 Guard rails against AI mistakes
Every number and date in a summary must appear in the episode transcript, or it's removed. Stock tickers are checked against Nasdaq's official list. A forecast only reaches the Scorecard if the speaker's exact quote is found in the transcript. Headlines and lessons that are vague or sound like advice get rewritten automatically.
⏰ The database is the clock
A scheduler inside the database (pg_cron) starts the price job every 15 minutes while the US market is open and the daily digest each weekday at 5:30 am New York time. It knows the NYSE calendar, so holidays and 1 pm half-days never fill a trade or score a call at a stale price.
🎮 Cheat-proof games
Paper trades and predictions are settled inside the database, never in your browser. Orders fill at the next market-hours price (so nobody can trade on old news), quiz scores are checked against hidden answer keys, and rewards can only be paid once per achievement.
📈 Forecasting science
Predictions carry a confidence level and are scored with the Brier score, the method from forecasting research, so leaderboards reward good judgement rather than lucky guesses. Your calibration chart shows whether "80% sure" really means 80%.
🧭 A coach that watches your habits
The database looks at your own trades for classic investor mistakes (panic selling, putting everything in one stock, overtrading) and links each one to the lesson that explains it.
🔄 Instant sync, works offline
Progress saves on your device first, so taps feel instant even offline. Signed in, it syncs to the cloud in about a second and appears live on your other devices. Conflicts between devices are merged automatically, so nothing is lost.
🔐 Private by design
Row-level security means each account can only read its own data, enforced by the database itself. Sign-in is a one-time email code, with no passwords stored by us. Usage statistics are anonymous.
📱 One codebase, every screen
The same site is a website, an installable web app (offline-ready), and Android and iPhone apps (native Capacitor shells). New features reach every phone without anyone reinstalling.
The full stack
- AI (cloud): Google Gemini Flash / Flash-Lite with automatic model fallback; Kokoro-82M text-to-speech (ONNX on CPU)
- AI (offline mode): Whisper large-v3-turbo (mlx-whisper), Qwen3 14B (Ollama), Kokoro-82M (mlx-audio). Either way, structured JSON outputs are validated by Pydantic
- Pipeline: Python, RSS (feedparser), Yahoo Finance data (yfinance), Jinja templates → static HTML
- Front end: plain HTML, CSS and JavaScript (no framework), hand-drawn SVG charts, a service worker for offline use
- Backend: Supabase (Postgres, auth, realtime) with row-level security and database functions for all game logic
- Hosting & automation: GitHub Pages, GitHub Actions (daily digest, 15-minute price job, deploys), started by Supabase pg_cron + pg_net, with the GitHub token in Supabase Vault
- Mobile: Capacitor 7 Android and iPhone apps; the web app also installs to any phone's home screen