Skip to content
Live demo
Live2026Solo founder, end-to-endEdTech, Consumer SaaS

LiraLang

An AI tutor for Italian, taught from Spanish — the 82% you already know, plus the 18% you don't.

Live demo
Course Dashboard
https://www.liralang.com
LiraLang course dashboard with chapter tree, streak/precision counters and skill radar
Open
0%
Vocabulary in common
IT–ES baseline
0
Chapters
A0 → C1
Proficiency span
Real-time
AI corrections
Why this exists

Italian and Spanish share roughly 82% of their vocabulary, yet mainstream language apps start every learner from zero and explain new concepts in English. Spanish speakers learning Italian waste months on cognates and structures they already understand, then hit a wall on the 18% that diverges — false friends, irregular conjugations, particles like ne and ci, gendered-article quirks. No widely-available platform meets bilingual learners where they actually are.

What it is

A web platform organised as 20 chapters across five CEFR levels (A1 → C1), with curriculum built around comparative grammar. Every concept is introduced through what is identical to Spanish first, then what diverges — not from scratch. Lessons surface cognates and structural parallels, then isolate the differences and drill them with targeted exercises.

The conversation surface is an AI tutor powered by Anthropic Claude through the Vercel AI SDK, streaming corrections live during dialogue and writing exercises. Mistakes are explained in the learner's native Spanish with grammar references and worked examples — not translated, but reasoned through. An adaptive spaced-repetition engine weights review by recency and error category, generating exercises from past mistakes rather than a generic schedule.

Learner profiles capture an archetype (explorer, perfectionist, speed-focused, writer, consistent) and self-flagged weak areas, feeding tone, difficulty scaling, and exercise selection. The tutor personalises within minutes of signup rather than after weeks of data collection.

Stack

Built with

A pragmatic stack — selected to ship fast without trading away long-term operability.

19 technologies · 6 layers

Frontend

7 tools

  • Next.js 16
  • React 19
  • TypeScript
  • Tailwind CSS 4
  • shadcn/ui (Base UI)
  • Framer Motion
  • React Markdown

Backend

3 tools

  • Next.js API Routes (App Router)
  • Prisma 7.7 ORM
  • pg adapter

Database

2 tools

  • PostgreSQL
  • Spaced-repetition schema indexed on error category

AI / ML

4 tools

  • Anthropic Claude (Haiku 4.5, Sonnet 4.6)
  • Vercel AI SDK
  • Real-time streamText
  • Comparative-grammar prompts

Auth & Security

2 tools

  • NextAuth v5
  • Email-credentials flow

Infra & DX

1 tool

  • Vercel
Highlights

What makes it work

01

Comparative grammar by design

Every concept teaches what is identical to Spanish first, then what diverges. Eliminates the months of redundant learning that generic apps impose on bilingual students.

02

AI tutor in the learner's native language

Claude-powered conversation and writing feedback catches mistakes live and explains them in Spanish — with grammar references, contextual examples, and follow-up drills, not a translation.

03

Error-driven spaced repetition

Review exercises generate from actual past mistakes, weighted by recency and error category. Difficulty adapts session-to-session based on learner archetype and recent performance.

04

Native-Spanish-speaker UX

Interface and tutor speak Argentine Spanish (voseo) by default, personalise tone to the learner persona, and resurface saved vocabulary in new contexts for natural reinforcement.