Overview
Producing a professional podcast, ad, or piece of branded audio is usually expensive: multiple tools, hired voice talent, weeks of turnaround. Wondercraft set out to be the Canva of audio, a place where anyone could produce broadcast-quality sound fast and affordably. After closing a $3M seed round in January 2024, they had a narrow window to prove it before the audio AI space got crowded.
We joined Wondercraft's team directly and shipped the core of their new audio studio. Ex: one-click audiobook generation, instant podcast translation, and lip-sync accurate enough for real human performance. One of the studio's first serious tests was El Daheeh, the Egyptian YouTuber known for fast, expressive delivery that most lip-sync models can't keep up with. We built one that could.
Audible signed on as a six-figure recurring customer. Wondercraft's ARR grew from $300K to $2M. The studio launched to #1 on Product Hunt.
Before
Wondercraft builds the tools that make producing podcasts, ads, and other audio content as easy as designing a slide deck. They believe audio creation shouldn't need a studio, a hired voice, or weeks of turnaround. Audio creation should be as accessible as Canva made design. Their MVP handled the basics: simple dubbing only for speech. It had no way to work with music, and the system underneath it for grouping speech into usable segments had been hacked together to get the MVP out the door not scalable.
The Call
In January 2024, Wondercraft closed a $3M seed round with a deadline of launching the new studio by January 15th, before the category got any more crowded. Their CTO, head of engineering, and a frontend engineer were already coding around the clock to get the fundamentals in place. What they needed was someone who could move at that same speed on the hardest part.
A recruiter connected us, and the they asked about concurrency. We talked through how Python handles parallelism under the hood and how we'd architect genuinely distributed processing for their scale. That plus the way we write code, LeanArchitecture™, made them confident we will help them hit the launch date.
How We Got There
We worked inside their team from day one with daily standups, a running channel directly with CTO Youssef Rizk for anything technical, a shared Notion board. We worked together as a very fast team that also found time to laugh, including the time Oskar Serrander, co-CEO, discovered the flip-up sunglasses clip on Muhamed's glasses and refused to let it go.
Beside the jokes we did full-stack, infrastructure, and machine learning, owned end to end, for two months straight, most days pulling 20 hours coding sessions. We hacked the dubbing pipeline to extract and group music separately from speech, extended the MVP's existing segment-grouping system rather than throwing it out, and rebuilt processing so it could actually scale.
What We Built
We helped produce an audio studio that does in seconds what used to take a production team days with one-click audiobook generation, instant podcast translation, and lip-sync accurate enough for real performers. One of the studio's first serious tests was El Daheeh, an Egyptian YouTuber whose face moves more than almost any generic lip-sync model can track. We trained a model specifically capable of following him.
We also fixed the problem of speed at scale. Before this, processing time in the studio scaled with the length of the audio, so a longer file simply took longer to process. We segmented and parallelized the pipeline so that stopped being true. Now, no matter how long the file is, processing takes seconds.
The Impact
Audible signed on as a six-figure recurring customer. Wondercraft's ARR grew from $300K to $2M. The studio launched to #1 on Product Hunt.
Want to hear our thoughts on accelerating engineering performance via AI? Watch to learn more.
