The Problem
Language testing is a $250, two-week bottleneck
Over 1.5 billion people are learning English globally, and 130 million are learning German. The demand for verified language proficiency is accelerating, driven by remote work, cross-border hiring, immigration requirements, and study-abroad programmes.
Yet the testing infrastructure hasn't kept up. IELTS and TOEFL cost an average of $250 per test, take 7–14 days for results, rely on manual evaluation that introduces bias, and simply cannot scale to meet global demand. For enterprises screening candidates across geographies, for EdTech platforms that need embedded assessment, and for students and migrants who need affordable certification. The current system is broken.
The convergence of LLMs (GPT-4o, Claude), speech-to-text AI (Whisper, Deepgram), and the CEFR framework created a window to build something fundamentally different: an AI-native assessment platform that delivers accurate, standardised results in minutes, at a fraction of the cost.
Market Opportunity
$30B market with a clear wedge
I sized the opportunity using a TAM → SAM → SOM framework:
- TAM: $30B: Global language learning and testing market (HolonIQ, 2023)
- SAM: $8B: Digital proficiency assessments (the slice where AI can replace manual testing)
- SOM: $200M: Initial wedge targeting B2B screening (HR teams, immigration services) and B2C certification (students, job seekers), informed by Duolingo English Test and Berlitz benchmarks
The market timing was right: enterprises were already searching for scalable screening tools, Duolingo's English Test had validated consumer willingness to trust AI-scored assessments, and LLM capabilities had crossed the accuracy threshold needed for CEFR-level evaluation.
User Personas
Three distinct users, one platform
The product needed to serve both sides of the market: the organisations that need assessments and the individuals taking them:
HR Manager (B2B): Mid-level talent acquisition at a multinational, screening candidates for language readiness in customer-facing or international roles. Pain: manual screening, interviewer bias, inconsistent evaluation. Need: fast, auditable, CEFR-aligned reports integrated into their hiring workflow.
Student / Migrant (B2C): Preparing for university admission or visa applications. Pain: $250 test costs, weeks of waiting, test anxiety. Need: affordable, reliable certification that institutions accept.
EdTech Platform (B2B API): Product team at a language learning startup that needs assessment capabilities without building them in-house. Pain: no scalable evaluation engine. Need: plug-and-play API, customisable branding, actionable performance analytics.
The product moat isn't the AI; it's the trust layer. Anyone can build an LLM-powered scoring engine, but getting enterprises and institutions to accept your scores requires transparent rubrics, human-reviewer calibration, bias audits, and GDPR compliance. The technology is table stakes; institutional trust is the differentiator.
Solution Architecture
From prompt to CEFR report in under 10 minutes
The product flow was designed around a simple loop: users respond to AI-generated prompts → speech and writing inputs are processed by GPT-4o and Whisper → a CEFR-level report is generated instantly.
The platform evaluated four skill areas (speaking, writing, listening, and reading) using CEFR descriptors (grammar, fluency, coherence) as the scoring rubric. Each assessment was designed to complete in under 10 minutes, with results available immediately.
Tech stack: React + TailwindCSS (frontend), Node.js / Python FastAPI (backend), GPT-4o / Claude for LLM scoring, Whisper / Deepgram for speech-to-text, Auth0 for authentication, PostgreSQL for data, AWS S3 for recordings, Mixpanel / PostHog for analytics.
Revenue model: B2C freemium ($10/test, $50 for a certified report) + B2B per-seat pricing + API licensing (per-session or flat fee). The pricing was designed to be 95% cheaper than IELTS while maintaining enough margin for the AI inference costs.