Everyone is learning a language, be it on their own, using an app, or with a teacher. We're learning Portuguese, taking two classes a week.
Having mastered the basics, our challenge now is to expand our vocabulary. We dedicated an evening to creating Anki flashcards, only to discover that the app wasn't free on the App Store, and its user interface was horrendous.
We discussed that we could build a solid dataset, with various AI models, in a few hours to make the ultimate flashcards. The ultimate flashcard would work from any language to any other, featuring a beautiful image representing the word—since visualizing the word is more effective than merely reading its translation—and an audio clip for pronunciation, as hearing a word can also be more beneficial than reading it.
We named our creation Unolingo, ultimately envisioning it as a companion to Duolingo, and because the cards reminded us of those in the Uno game. More importantly, It's easily searchable on the App Store, which will help with our social sharing strategy experiments.
Our initial project was a 24-hour hackathon, which we completed successfully, resulting in a live app on the App Store.
We initially designed category card packs, a concept we abandoned in the following update, shifting to a limit of seven new words per day. We compiled over 2000 categorized words, translated into 19 languages - including pronunciation - using GPT-4, and generated over 20,000 multilingual audio samples.
Sourcing images was challenging. As we were not willing to spend 4 cents per image, we found a hack around. We crafted prompts for DALL-E 3 via ChatGPT, generating about 50 images of the style we were after. These served to pre-train an SDXL model on Replicate, enabling us to produce thousands of images at a fraction of DALL-E's cost. Despite occasional lower-quality output, this approach drastically reduced expenses.
For distribution, we chose to immediately release the app in the 19 languages we supported. Originally named 'Language Vocabulary Flashcards' for App Store Optimization (ASO). Localizing the app involved translating the name, description, and all store screenshots. This strategy generated over 2000+ store impressions and 200+ downloads in the first few days, leading to four paid subscriptions overnight.
The initial version, with its cumbersome packs and imperfect UI, particularly the paywall and some user interactions, needed refinement. So, we focused on enhancing these aspects, rebranding the app as Unolingo, and introducing a new, more shareable concept of learning seven new words daily. We included shareable cards, reminder notifications, and ads.
(To be continued...)