GoJo

• 

  Landing Page

Inspiration

GoJo started with a simple frustration: personalized news products often create more work. They give users endless feeds, repeated headlines, and AI-generated summaries that say a lot without delivering useful information.

We wanted to build something closer to a personal radio news desk: short, fast, source-backed audio alerts covering exactly what the listener cares about.

What it does

Users choose three main topics, select relevant subtopics, and rank their preferred listening order. GoJo researches recent reporting and creates a personalized audio rundown.

Each alert:

• Covers one specific story
• Names the reporting source
• Includes a useful detail
• Avoids repetition and filler
• Uses different natural-sounding voices

How we built it

GoJo uses:

• OpenAI for web research and structured alert generation
• ElevenLabs for natural, multi-voice narration
• JavaScript, HTML, CSS, and Node.js
• Vercel for hosting and serverless API routes

The system validates sources, removes unsupported alerts, and checks semantic similarity to prevent the same story from appearing under overlapping topics.

Challenges

The biggest challenge was controlling AI verbosity. Early versions produced long, podcast-style explanations that sounded polished but communicated very little.

We redesigned GoJo around strict editorial rules: one event per alert, source attribution, concrete details, hard word limits, and no forced commentary.

Source reliability was another challenge. Models can generate plausible-looking URLs, so GoJo now uses links returned by the search system, verifies them, and removes alerts that cannot be connected to a trusted source.

What we learned

We learned that better AI news does not necessarily mean longer summaries. Concision, attribution, and editorial restraint make AI-generated audio significantly more useful—and make synthetic voices feel more natural.

The strongest version of GoJo is not an AI podcast. It is a personalized audio newswire: more interests, fewer words, and a much higher signal-to-noise ratio.

Built With

• css
• elevenlabs
• html
• javascript
• node.js
• openai
• responses
• vercel