Deep Dive: Building Voxera and What's Next

In partnership with

Hey there,

Last weekend, I built Voxera - a tool that turns any blog post into a podcast episode.

Three friends signed up within hours of me sharing it.

Here's exactly how I did it and what you can learn from the process.

The Problem-First Approach

I picked a problem I had. 

I wanted to listen to blog posts during walks but hate robotic text-to-speech.

Lesson: Start with your pain points. If you're frustrated by something, others probably are, too. I could have built another ToDo app, but I wouldn't use it myself.

I spent Friday night mapping out the flow on paper. 

URL input → scrape content → AI summary → natural speech → downloadable MP3.

Why this matters: Paper beats code every time for initial planning.

I spent 30 minutes drawing boxes and arrows instead of coding for 30 hours in the wrong direction. This simple flow became my north star.

The Technical Architecture

I started with the scraping. I used Firecrawl API because I didn't want to fight with different blog formats and paywalls.

Key insight: Don't reinvent wheels that others have perfected.

Firecrawl handles JavaScript-heavy sites, removes navigation, and deals with paywalls. Building this myself would have taken weeks.

API cost: FREE tier. $0

Turn Creative Briefs into Revenue

With Neurons AI, agencies can fast-track everything from creative development to client sign-off.

Neurons AI indicates how campaign assets will perform—before launch—so you can deliver smarter, faster, and with confidence.

Get instant insights on attention, engagement, and memorability, along with AI-powered recommendations to refine assets before clients even see them.

No more endless feedback loops. No more debating design choices. Just clear, data-backed reasoning that helps you present and sell creative work with ease.

Using Neurons AI, Teads helped their client improve brand awareness by 64% and ad recall by 29%.

Get Your Neurons Demo

I added OpenAI for summarization. Long blog posts become tight 3-5 minute summaries that make sense when spoken.

Technical detail: I used GPT-4 with a custom prompt that optimizes audio consumption.
Written content and spoken content have different flows. I had to train the AI to create summaries that sound natural when read aloud.

Cost: $0.03 per 1000 tokens. The average blog post costs $0.15 to summarize.

I used ElevenLabs for voice generation. Their API turned my summaries into podcast-quality audio that doesn't sound like a robot.

Why ElevenLabs over alternatives: Their voices have emotion and pacing.
Google's TTS sounds robotic. ElevenLabs costs more but the quality difference is massive.

Cost: $0.22 per 1000 characters. A 5-minute podcast costs about $1.50 to generate.

The Frontend Strategy

I built the frontend with Next.js and Tailwind.

Clean, dark theme.

One input field.

Big convert button.

That's it.

Design philosophy: Complexity kills conversion.

Instagram started as a simple photo filter app.
I could have added 20 features, but users want to paste a URL and get audio. Everything else is noise.

I added Clerk for auth and Supabase for storage.
Users can save their conversions and download them at a later time.

Technical choice: These tools handle the mundane tasks (authentication, database management, file storage) so I can focus on the core feature.
Clerk saves you weeks of authentication implementation.

Supabase provides PostgreSQL with real-time features and built-in file storage.

The Validation Process

I called it Voxera.

I tested it on 20 different blog posts.

It worked every time.

Testing strategy: I picked posts from different platforms.

Medium, personal blogs, news sites, technical documentation.
Each had different HTML structures. Firecrawl handled them all, but I learned which content types work best.

I added customization options, including different voices, speech speeds, and intro music. People want control over how their content sounds.

User feedback insight: My first version had one voice and fixed speed.
Beta users immediately asked for options.
People want their audio to match their brand, even for personal use.

The Business Layer

I set up basic billing with Stripe. Free tier gets 5 conversions per month. Paid users get unlimited.

Pricing strategy: Free tier proves the value. $9/month for unlimited feels fair when each conversion costs me $2 in API calls. The math works at scale.

I deployed on Vercel Saturday night. 
Domain, SSL, CDN - all handled automatically.

Deployment insight: Vercel makes deployment feel like magic. Git push triggers automatic builds, SSL certificates, global CDN deployment. No server management, no downtime worries.

The Launch

I shared it with 10 friends Sunday morning.

3 immediately signed up and converted their blog posts.

Launch lesson: Start with friends who will give honest feedback. They found bugs I missed and suggested features I hadn't considered. Their enthusiasm told me I had something worth building.

What's Next for me:

I’m picking up another idea I have had for a while now, and complete the build over the weekend. Will share everything details about it with you next week.

The Real Takeaway

The magic wasn't in the code.

It was in picking the right problem and the right tools.

I used to think "real developers" build everything from scratch.

Now, I think "smart developers" glue together great APIs and focus on user experience.

Voxera works because it solves one problem well.

I listen to any blog post on my morning walks now.

Sometimes, the best weekend projects are those that address your daily frustrations.

What problem are you going to solve next weekend?

Building in public, John

P.S. - Want to see Voxera in action?

Reply to this email, and I'll send you early access.

Still working out the kinks, but it's functional enough to impress your friends. If this was helpful, forward it to someone building their weekend projects.

They'll thank you for it.