AI Jemmel

AI Jemmel turns doom-scroll into a witty 3-minute audio digest—daily or on-demand—so you’re informed, not overwhelmed.

Comment

Inspiration

Five days ago I was perched on my in-laws’ couch, thumb locked in a doom-scroll trance following the Iran–Israel conflict.
The living-room conversation kept rolling, but I was mentally nowhere near it.
That jarring disconnect sparked a question:

Could news be delivered in a calmer, connection-friendly way?

I had already registered for the Bolt.new Hackathon with a different (fairly generic) idea.
That night I scrapped it, opened a fresh Bolt workspace, and sketched Calm Feed—the seed that would grow into AI Jemmel.

What it does

AI Jemmel transforms thousands of headlines into a witty, three-minute audio digest you can finish while the kettle boils.

  • Daily Digest – Press play and get all major stories, anxiety-free.
  • One-Topic Generator – Signed-in users type any topic (“AI ethics”, “SpaceX”, “K-pop”) and get a custom mini-show.
  • Pro Playlist – Power users stack multiple interests into one seamless audio clip.
  • Click-through Context – Every segment sits atop its source article link—one tap for deeper reading, no rabbit holes.

How we built it

Day Milestone Stack
-5 Doom-scroll epiphany at in-laws → pivot to news calm-tech Notes in Notion
-4 Bolt.new project scaffold “Calm Feed” Bolt Starter, Supabase Auth
-3 Ran out of Bolt credits → switched to local dev in VS Code Next.js, Tailwind
-2 Added ElevenLabs TTS for narration ElevenLabs API
-1 Prompted Google Gemini to write late-night-show scripts → “Jemmel” persona born Gemini Pro
0 Polish UI, gradient icon, shoot demo, Devpost copy Vercel Deploy

Challenges we ran into

  1. Credit Crunch – Free Bolt quota vanished fast; rebuilding and caching summaries kept costs at $0.
  2. Tone Tuning – Scripts sounded either too stiff or too slap-stick; dozens of Gemini prompt tweaks finally nailed “witty but trustworthy.”
  3. Latency – Chained calls (scrape ➜ GPT-4o ➜ Gemini ➜ ElevenLabs) felt slow; streaming TTS and memoizing summaries cut wait time to ~6 s.

Accomplishments that we're proud of

  • Concept-to-production in < 5 days while traveling.
  • First share with non-tech friends got a genuine “Wow, this is fun—and actually useful.”
  • End-to-end serverless: Bolt frontend & API, GPT-4o summaries, Gemini scriptwriting, ElevenLabs voice, Vercel hosting.
  • Created a cool intro music for AI Jemmel As well.

What we learned

  • Constraint breeds creativity—running out of credits forced smarter caching and leaner APIs.
  • Audio UX > text for emotional resonance; people remember a calm voice far better than a wall of headlines.
  • Bolt.new’s rapid scaffolding changes how quickly an idea can leap from spark to shipping product.

What's next for AI Jemmel

  • Realtime Spike Alerts – Optional push when a followed topic suddenly trends.
  • Mood-Adaptive Voice – Softer tone for heavy news, upbeat for science breakthroughs.
  • Wearable & Smart-Speaker Skills – Hands-free listening on daily commutes or while cooking.
  • Community Playlists – Curated mixes by journalists, educators, and—why not—late-night hosts themselves.

Built with ❤️ on Bolt.

Built With

  • bolt
  • elevenlabs
  • gemini
  • nextjs
  • supabase
Share this project:

Updates