How a 19-year-old built a $1.5M AI audio SaaS in seven days

Executive overview

Arab, a self-taught developer, spotted a Discord of non-technical musicians struggling to run open-source vocal conversion models locally. He built a simple web wrapper in a week, launched it, and hit 100,000 users in the following seven days.

The product — MusicFi — grew to nearly 2 million users and over $100K monthly recurring revenue within eight months. A cease-and-desist from Universal, Sony, and Warner forced a pivot away from celebrity voice models toward custom voice training, which accidentally created a stickier, more defensible product. The company is now transitioning toward B2B audio tooling for gaming studios.

The fastest path to product-market fit is launching the roughest version into a community that already wants it, then iterating in the wild.

From idea to 100,000 users in two weeks

• Discovered a Chinese research paper on vocal style transfer; noticed non-technical users trying to run it manually in a Discord called AI Hub
• Built a minimal site in one week — three celebrity voice options, one upload button, no authentication
• Announced the launch to the Discord founder; the community drove the first wave of traffic
• Costs hit $3,000/day in GPU credits before any monetisation was in place
• Added a $1/month paywall after spending hit $10–12K; made $2,000 on day one
• Raised pricing iteratively; the constraint of mounting GPU costs forced monetisation decisions

The cease-and-desist pivot

• Three weeks after launch, received simultaneous letters from Universal, Sony, and Warner Music
• Removed all celebrity voice models the same day
• Replaced them with 100 synthetic voices created via linear interpolation — blending model weights from two real voices to produce a new voice with no copyright owner
• Reframed the product around professional music use cases: vocalists converting their voice to different timbres, producers generating demos in an artist's style without hiring them
• This pivot attracted higher-retention professional users, including named clients like Louis Bell (Post Malone's producer)

Distribution strategy

• Celebrity AI covers generated organic controversy and viral sharing before the product even launched publicly
• Built an affiliate program using Tolt.io: anyone could generate a link in under five minutes, earn 25–30% of referred revenue
• ~4,000 affiliates created TikTok pages and posted AI cover content with affiliate links embedded
• Identified a key insight: TikTok traffic converts poorly (entertainment intent); YouTube long-form converts well (producer intent) — one 300K-view YouTube video drove $30K in revenue in a month
• Viral demo format: record your own voice, convert it to a celebrity voice live on screen — hooks viewers instantly

Moving upmarket: B2B audio for gaming

• Music software TAM is under $1B; gaming audio TAM is ~$250B and expected to double
• Gaming studios pay $50K+ per game to sound agencies; MusicFi can replace that with an in-house generative tool at ~$30K/year
• Planned product lines: voice actor replacement, generative music for game soundtracks, text-to-sound-effects via LLM
• Strategic investors chosen for access rather than capital: Jaroslav Bek (founder of Beat Saber, sold to Meta for $2B), and Founders Inc
• B2B contracts are enterprise-scale, high-retention, and provide an exit multiple that consumer SaaS cannot

How to find trends before they become obvious

• Read new submissions on arXiv.org daily — research papers surface trends weeks or months before products appear
• Hugging Face shows which new models are gaining traction among developers
• Hacker News aggregates the most discussed technical developments
• The signal: a research paper exists AND early organic content is already spreading — that combination means a market is forming

How to launch a SaaS without code or capital

• Distribution is the scarcest resource, not code — find a creator or community with an audience first
• With distribution locked in, approach developers on Discord (AI Hub, Founders Inc), GitHub, or Hugging Face; developers want users, not money
• Drop a technical problem in a developer chat; whoever solves it cleanly is worth recruiting
• Build a community or themed social account around the problem before the product exists
• Affiliates are underused: provide a shareable link with a financial incentive and let existing content creators do the marketing
• Create a demo video that lets viewers imagine themselves using the product — the "I sound like Drake" format

Tech stack and internal tools

• TypeScript / JavaScript with Next.js; Supabase for the database
• Modal (serverless GPU hosting) instead of fixed GPU rentals — spins compute up and down with demand, eliminating waste
• Closed-source audio-to-audio API available to B2B clients at a $10K minimum spend
• PostHog — session recordings, heat maps, user analytics; one line of code to install
• Retool — internal dashboards built on top of the database
• Paddle (not Stripe) — adds PayPal acceptance, drove a 20–30% increase in subscription conversions
• Tolt.io — affiliate link management, full dashboard for tracking referrals and conversions
• Customer.io — CRM and email automation
• Superhuman — email; cited as the canonical example of a product people would "fight to get back"
• Cron — calendar

Principles on building and scaling

• "Iterate in the wild" (GTFL — get the fuck off localhost): launch the first version before authentication, rate limiting, or polish exist
• When given two options, ask why you cannot do both before committing to one
• Retention problems at scale are harder than growth problems — a single UI change can move conversion ±5%
• Understanding your own product at a low level (not necessarily writing the code) is the founder's job; founders who cannot explain their own architecture cannot debug it at scale
• Consistency compounds: Arab went through a clothing brand, a chocolate resale, a stock-trading Discord, and four other apps before MusicFi
• Success formula (his framing): 80% consistency, 10% luck, 10% skill