DEV Community

Cover image for From Frustration with News to an AI Radio Startup (with Full Stack Breakdown)
Mike
Mike

Posted on

From Frustration with News to an AI Radio Startup (with Full Stack Breakdown)

At some point, I wanted to dive into Kubernetes and mobile development, but I wasn’t sure where to begin. At the same time, I was deeply frustrated by the lack of news free from political agendas and negativity. The internet felt flooded with manipulative content, and I found myself wishing for a filter that could cut through all the noise.

Eventually, I started by creating my own AI-powered online radio that broadcasts copyright-free music and what I consider “clean” news — free from bias and negativity. Later, when people began reaching out, asking how to create their own internet radio stations, I decided to focus my efforts on building a full-fledged platform with AI hosts, podcasts, and an intuitive management system.

I ended up calling the project Tunio, from “tune” and “I/O”, but happy to hear if it evokes something else to you.

Streams that demonstrate the overall result

Quick breakdown of the stack I used

Stack Overview (Kubernetes-based):

  • Audio: Icecast2, Liquidsoap
  • Logic: Golang (real-time playlist management and background conversion jobs)
  • Storage: PostgreSQL (pgvector), S3
  • Processing: ffmpeg for news and podcasts — used for normalization and preparing all audio files before streaming.
  • Monitoring: Prometheus + Grafana
  • TTS: ElevenLabs, Piper TTS (self-hosted)
  • Frontend: Next.js, React Native (Expo)

TTS / Audio Generation

Originally used ElevenLabs, but later started using self-hosted Piper TTS to avoid high costs in the early stages of the project and to stay unlimited in content generation.

News and podcasts are generated from RSS → filtered from negativity, politics, ads, and promotions using AI models → deduplicated with vector embeddings → summarized using GPT → voiced → sent to the live stream.

Content

  • All music is copyright-free, generated or licensed via open models
  • News is categorized (tech, sports, gaming, etc.) and updated frequently from RSS-feeds
  • Jingles are also generated using TTS, with voices provided by ElevenLabs.
  • Podcasts are generated based on user prompts, voiced by the selected TTS provider, and sent to the broadcast.

Frontend

  • Next.js with next-intl for i18n for application and control panel
  • React Native (Expo) app with WebView for mobile access

Distribution

Built-in restreaming to YouTube, Twitch, and Telegram directly from the platform — no third-party software needed.

I’ve set up a few demo stations to test different use cases (lounge, ambient, etc.) — would be curious to hear what you think: https://app.tunio.ai

I’m not able to fully onboard users yet, as the company registration process is still underway and I can’t accept payments at this stage — but the platform is open for testing.

If anyone’s interested, I’d be happy to share my story — the mistakes I made early on and how I had to rethink and redo some of the initial decisions.

Feedback of any kind is welcome — tech suggestions, naming thoughts, or just impressions!

Top comments (0)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.