2024 / case study

SongSens.ai

Learning a language through the music you already love.

making language learning sing.

A language-learning platform built around the songs you can't stop replaying. Dynamic translation, phonetic guides, and word-by-word context, shipped solo on a fine-tuned LLM stack.

YEAR

2024

ROLE

Product · UI/UX · Full-stack

TEAM

Solo

STACK

Svelte · Node · AI & LLM

LAUNCH

Product Hunt · 2024

STATUS

Live

visit the site

the problem.

I wanted to learn Chinese through music. Every translation tool I tried gave me flat literal lines that stripped the culture, the slang, and the sound of the language. AI was just emerging, and I had a hunch it could be the tutor that didn't yet exist.

the process.

Every step went live before it was polished and got reshaped by what real users actually did with it. The brand, the design, even the core flow changed after launch, not before.

Test models, tune prompts

Compared several frontier LLMs across thousands of lyric snippets. Picked the model that read slang and cultural context, not just literal meaning, then iterated the prompt structure until cost-per-translation was low enough for a freemium plan to work.

Design the lesson flow

Sketched the interface around one rule: AI in the background, lyrics in the foreground. Translation appears automatically; deeper context (slang, grammar, pronunciation) unfolds on tap. No chatbox, no assistant monologue, no settings to wade through before the first lesson.

Draw the brand

Mascot, type system, colour palette, motion language. The identity needed to carry warmth so the product would read less like a translation tool and more like a tutor you'd actually want to spend time with.

Build the full stack

Shipped it solo: Svelte front-end, Node back-end, Postgres for users and lessons, auth, Stripe checkout, and the LLM prompt orchestration on top. Coming from design, owning the back-end was the steepest part, and the most rewarding.

Launch and learn

Product Hunt launch day, paid ads to test acquisition, friends-and-family beta to stress-test the lessons. Thousands of visits in 24 hours and first paying customers within the launch week, then on to learning what actually kept people coming back.

Iterate from the signal

After launch every change came from somewhere real: usage analytics, support emails, conversations with users. The mascot, the warmer tone, the simpler design, none of those were on the original roadmap. The product I shipped first was deliberately rough; listening to it in use reshaped it more than any planning round could.

from simple to a real mood

build first, fine-tune later.

The first version shipped clean and functional, but cold: a translation utility with no personality to hang on to. Real users told me what was missing: warmth, character, a face. The redesign was the answer to that feedback: a mascot, warmer colour, and a tone that made the product feel like a tutor you'd actually want to spend time with.

before

after

Homepage: from a quiet utility to a mascot-led welcome.

before

after

From a plain search field to the new design language carrying the same flow.

launched, learned.

SongSens.ai launched on Product Hunt, pulled thousands of visits in a day, and converted its first paying customers within the week.

Pulled 1,000s of visits in 24 h

Product Hunt + organic.

Converted its first paying customers

Within the launch week.

Launched on Product Hunt

Solo launch, 2024.

the honest part

The fit wasn't quite right. Most arrivals wanted audio-to-text translation, not lyric-deep language learning. It runs quietly in the background now. What it taught me doesn't.

I set out to learn Chinese through music and ended up learning how to launch a product solo: pricing, positioning, prompts, payments, ads, and the gap between a thing people will visit and a thing people will keep using. Worth every hour.

view website for this case

next case

Charlie's Clinic.

Turning a dental clinic into somewhere you actually want to go.

read case →