Word Radar.

The problem

Intermediate language learners hit a vocabulary plateau. They consume authentic content — podcasts, YouTube, TED talks, conversations — but encounter unfamiliar words constantly with no frictionless way to capture, translate, and retain them in context. By the time they reach a dictionary, the moment and the sentence are gone.

Existing apps teach fixed word lists disconnected from what the learner actually consumes, offer no awareness of personal vocabulary gaps, and require dedicated study sessions that compete with listening time. Flashcards strip words from their original context, hurting retention.

The solution

WordRadar is a mobile-first AI vocabulary coach that turns the words a learner encounters in real content into a personalized study curriculum. During onboarding it estimates the user's CEFR level from unknown-word taps in a reading passage, then builds a personal word bank.

Its core differentiator is passive ambient listening with personal vocabulary gap detection: it runs in the background while users consume content and surfaces only words genuinely unknown to that specific learner, each shown with its exact original context sentence. Users can also paste a YouTube, TED, or podcast URL to extract vocabulary asynchronously, and bias suggestions toward a chosen public figure's lexical style.

How it works

Live mic or imported audio is transcribed by Whisper; the transcript, the user's profile (target and native language, CEFR level, and up to 500 recently confirmed known words), and instructions are passed to Claude Sonnet, which identifies unknown words and returns strict JSON word cards — word, translation, exact context sentence, and a level-appropriate usage note. The API is called only server-side from an Express.js backend, with Supabase storing per-user vocabulary profiles.

Guardrails require verbatim context sentences, suppress obvious cognates for English speakers learning Romance languages, filter to the target language, and gate on Whisper confidence to avoid analyzing noisy audio. Prototype testing across five runs reached ~93% overall, with JSON compliance at 100% and known-word filtering at 100%; the main improvement areas are verbatim context enforcement and contextual translation. RAG (OpenAI embeddings) powers speaker-style mode by retrieving from public-figure corpora.

Who it's for

WordRadar is B2C, built for "The Immersion Learner" — a self-motivated adult aged 25–45 at intermediate level (A2–B2) who consumes foreign-language media and wants efficient vocabulary growth without structured study. They are time-constrained (30–60 minutes a day) and prefer learning integrated into existing habits.

Revenue is freemium: a 14-day free trial converting to a monthly subscription (~$9.99/month), where the free tier stops scanning but preserves word history to encourage reactivation. Secondary post-launch monetization includes an annual plan and B2B2C licensing to language schools and corporate L&D.

Why it matters

The global language-learning app market was valued at $6.34B in 2024 and is projected to reach $24.39B by 2033 (~16% CAGR), with the AI-powered segment growing fastest. No competitor passively listens to ambient audio and cross-references a personal vocabulary profile in real time — WordRadar's claimed uncontested space, against incumbents like Duolingo, Babbel, LingQ, and Langua.

Currently a pre-seed 0-to-1 startup with a working prototype, the launch is a phased soft launch: a 50–100 user closed beta, then an App Store soft launch with a 14-day trial targeting ≥15% trial-to-paid conversion, then full launch aiming for 500 paid subscribers within 60 days. Retention targets include D7 ≥40% and D30 ≥20%.

At a glance

Project

Word Radar

Built by

Alexey Yalyshev

One-liner

Word Radar is a mobile-first AI vocabulary coach that turns the words a learner encounters in real content into a personalized study curriculum.