AI French Tutor WhatsApp Bot
Building a zero-friction conversational AI tutor in WhatsApp that corrects grammar, logs vocabulary, and simulates TCF exam tasks using Node.js and Claude.
My Role
Architecture & Deployment
1 Week (Rapid Sprint)
Tech Stack
Node.js, Claude API, Twilio, SQLite, Railway
Zero-Friction Daily Practice
Challenge
Learning a language for a certification like the TCF requires constant practice, but traditional methods create unnecessary friction. Relying solely on human tutors is expensive, and standard apps lack conversational depth. The goal was to eliminate the barrier to entry for daily practice:
High Friction: Opening dedicated learning apps breaks the natural flow of daily communication, leading to inconsistent practice habits.
Lack of Context: Standard bots fail to provide structured, plain English explanations for specific grammatical mistakes in real time.
Manual Tracking: Pausing to write down new vocabulary during a conversation completely disrupts the learning experience.
Missing Exam Focus: General AI lacks the specialized coaching required to generate, grade, and evaluate B1 to B2 expression écrite tasks.
Process
Mapped the entire conversation loop to ensure high quality learning without sacrificing response speed or infrastructure security.
Architecture Design: Built an Express web server to act as the front door, handling Twilio webhooks and routing commands like vocab, exam, and reset.
Security & Verification: Implemented Twilio cryptographic signature validation to ensure only authorized WhatsApp traffic can interact with the server.
Prompt Engineering: Designed a highly structured system prompt for Claude to enforce level matched French, structured corrections, and hidden data blocks.
Infrastructure & CI/CD: Deployed a persistent SQLite database mounted on a Railway volume, backed by a GitHub pipeline that auto deploys new code pushes in under two minutes.
$6
Monthly Running Cost
100%
Automated Vocab Tracking
0
Downtime
Results
I architected a fully integrated AI tutor that lives natively inside WhatsApp, turning standard texting into a comprehensive and highly personalized language learning loop.
Instant Orchestration: A Node.js webhook service bridges the Twilio WhatsApp API and Anthropic's Claude to deliver near instant conversational replies.
Automated Vocab Logging: The system prompts Claude to invisibly tag new words, which the server parses out and saves to a persistent SQLite database for later review.
Exam Simulation: An integrated exam command generates specific B1 to B2 writing tasks and grades them just like an official examiner.
Actionable Corrections: Every message is analyzed for errors, providing consistent cross out to checkmark corrections alongside plain English explanations.
