dialectio.xyz

Learn Italian, French or Portuguese faster by leveraging the words you already know. Dialectio taps linguistic overlap so you can start speaking from day one.

Comment

Inspiration

Languages and code have always pulled me in the same direction. While studying both, I realised every new tongue I picked up was easier than the last, not because I was smarter, but because the previous ones kept leaving lexical breadcrumbs behind. Spanish and Romanian let me read Italian almost at first sight; with a few deliberate conversations the language clicked and never left. I looked for an app that treated that overlap as a feature, not as noise, and found none. Dialectio is my attempt to automate that “already-halfway-there” feeling for anyone who speaks a Romance language and wants to reach another.

What it does

At sign-up the learner declares a source language. From that moment every exercise, hint and AI dialogue is generated to exploit cognates and shared syntax first, and introduce real novelty only when it adds communicative power. A micro-exercise teaches the shape of a phrase, a short role-play tests it in context, and both steps feed fresh data back into the next prompt so practice stays balanced between familiar terrain and new ground. The result is momentum: you speak sooner because you never start from zero.

How we built it

Dialectio runs entirely in the Bolt.new IDE with a React + TypeScript front end and Tailwind styling. Supabase provides authentication, row-level-secure storage and edge functions. Conversational logic is orchestrated with LangGraph; language understanding and hint generation come from OpenAI embeddings; text-to-speech replies use ElevenLabs. For the public demo every costly LLM call is served from a cache layer so judges experience real speed without incurring tokens.

Challenges we ran into

Version control became the first hurdle: building inside an online IDE is fast, but keeping Git history clean while the database schema kept evolving required strict discipline. A second challenge was self-restraint, saying “no” to tempting features so the MVP stayed loyal to the original spec. Finally, adhering to a coherent design system in Tailwind forced me to rethink components I had already coded; every colour and spacing token had to pass a consistency test before shipping.

Accomplishments that we're proud of

The highlight is a ranking algorithm that predicts which Italian (or French, or Portuguese) words a Spanish-speaking learner will probably grasp at first sight. By mixing frequency data with cross-lingual embeddings the system surfaces cognates early and defers low-yield vocabulary, giving the user an immediate sense of progress.

What we learned

This sprint was a crash course in realistic time-boxing: features that did not amplify the core value were postponed without remorse. I also learned to transpose academic linguistics into executable rules, and to weave modern AI tooling, LangGraph orchestration, OpenAI embeddings, ElevenLabs TTS, into a single Bolt workspace without losing performance.

What's next for dialectio.xyz

The roadmap is clear: extend the engine to other language clusters, add persistence so learners can save external materials, and train an accent-feedback module that flags phonetic drift in real time. The end goal is a polished B2C product, building technology stack robust enough to license to fellow founders who share the vision or joining a startup where I can build a product like this.

Built With

Share this project:

Updates