At the July 14 VibeCoding meetup, Mike Onslow, CTO of Clarity Voice and co-host of the Artificial Antics podcast, walked us through a passion project that merged voice, AI, and rapid prototyping into one powerful showcase.
His goal?
To build a tool that could read articles aloud, summarize content using AI, and sync audio narration with on-screen text â all built with the curiosity of a weekend hacker and the discipline of an enterprise engineer.
đ The AI Article Reader
Mikeâs project accepted a URL, scraped the content with BeautifulSoup, summarized it using Gemini 2.5 Pro, and then generated spoken narration using 11Labs, complete with:
He used:
Cursor and Replit for AI-native development
Bubble to redesign and polish the UI
Flask, Celery, Redis, and PostgreSQL for backend logic
Gemini + 11Labs for summarization and text-to-speech
âI handed the code to Gemini and said, âDocument this.â It gave me a full architecture map and reintroduced me to my own app.â
đ§ From Code to Re-Discovery
After stepping away from the project for a month, Mike used AI to regenerate full documentation, including diagrams and a table of contents â enabling him to re-onboard himself without re-reading every line of code.
He emphasized the importance of:
Version control for AI-generated projects
Building modularly with backup and rollback in mind
Learning by iterating, not over-planning
âIâm looking at code less. But Iâm directing more. Thatâs where weâre headed.â
đ¸ The Cost of Voice
As someone building AI communication tools professionally, Mike gave a grounded warning:
âText-to-speech and speech-to-text are always the most expensive AI services â even if you self-host.â
He praised 11Labsâ quality but noted it quickly burned through free credits, leading him to explore more cost-effective alternatives for future builds.
đ Connect with Mike
Learn more about Mikeâs work at Clarity Voice, or tune in to Artificial Antics, where he and his co-host debate the cultural and technical realities of AI in todayâs world.