0 to 1 - AI Assistant
I’m a Brazilian learning French, and most of what I need to say depends on the situation: a comment in a design review, an email to someone I barely know. The tools I used didn’t let me set that context once and keep it.
Saving what I learned was worse. I’d ask, collect the phrases I wanted, export them into a file a flashcard app could read, and pay for that app just to practice them.
I designed Lexy as an AI language assistant that goes beyond translating a sentence. It considers who you are speaking to, what you are trying to say, and how formal the situation is, then suggests natural French that fits that context. A phrase for a design review should not sound like a message to a close friend.
I renamed it Lexy because I wanted more of an emotional connection: a short, friendly name tied to language. The mascot is a greyhound running with a roll of parchment in its mouth.
In Lexy a context is saved: a name, a short description of the situation, Casual, Neutral or Formal, and Spoken or Written. You switch contexts from a menu, and every answer is ranked for the active one.
Each phrase gets two scores, “In general” and “In this situation.” A sentence full of design jargon scores low in general but can score high in a design meeting.
You type what you want to say in your own language, and Lexy suggests natural French for the active context. In the recording, “I think this is perfect” becomes “Pour moi c’est nickel”—a conversational option for a design discussion rather than a word-for-word translation.
I kept the main answer easy to find, with alternatives available when people want another way to express the idea. They can compare options and open a phrase for more detail without leaving the conversation. The goal is a useful answer first, with room to explore rather than a wall of translations.
A translation is more useful when you understand the choice behind it. Opening a phrase reveals a pronunciation guide, a literal back-translation, and usage notes. A separate explanation connects the phrasing to the active context.
The recording uses “I really like the spacing here” in a design critique. Lexy explains the design vocabulary and relaxed tone, while the “In general” and “In this situation” scores distinguish everyday usage from suitability for that particular setting. People can then compare an alternative instead of treating the first answer as the only correct one.
When an answer is worth keeping, you can pick up the phrase and drag it straight onto a deck in the sidebar. Saving becomes part of exploring the answer, rather than a separate job of copying text, exporting a file, and importing it into another app.
This absorbed another concept I had, Kuori, about learning the vocabulary that matters to you. In Lexy that vocabulary comes from your own sentences and contexts.
Opening a deck turns the phrases you saved into a stack of flashcards. Each card starts with what you want to say; flipping it reveals the French phrasing, with pronunciation, meaning, and usage notes available on the answer side.
The recording moves from “Could you repeat that?” to its French answer and then through the next cards. Translation and practice stay in the same tool, so the language you needed in a real situation becomes the language you come back to learn.
The gamification starts with a meaningful goal: saying something you actually need to say. Trying different phrasings and getting explanations creates a loop of experimentation and feedback, connecting to Octalysis’s Empowerment of Creativity and Feedback. The aim is to help people understand their choices, not just copy an answer.1
Saving useful phrases into personal decks brings in Ownership and Possession. Practicing those phrases supports Development and Accomplishment, with progress grounded in language people want to use. The motivation comes from building a useful personal vocabulary rather than collecting points for their own sake.1
The early versions ran on mock data. With real translations an answer took more than 20 seconds, and moving to the Claude API only brought it down to 17. So I changed the order: the best phrasing streams in first and the alternatives follow one by one, so the first phrase now appears in about 4 to 5 seconds.