AI tutor

s an educator working at a Dutch VMBO (pre-vocational secondary education) school, I’ve seen firsthand how students—particularly newcomers with limited Dutch proficiency—struggle to keep up. This inspired me to create an AI-powered tutor that provides personalized, language-sensitive support and makes learning more inclusive and accessible.

The AI Tutor is a web-based platform where students can interact in real time with a 3D avatar that speaks their native language or Dutch, depending on their proficiency level. The application includes a dynamic vocabulary list, live voice-to-text translation, tailored language output options, and user data storage. It supports multilingual communication, tracks language skill development, and creates a safe space for independent learning.

The frontend was developed using Bolt.new and is connected to Supabase for authentication and project structure. User data—including name, native language, and language level—is stored in Airtable. Voice input is transcribed using OpenAI’s Whisper API, translated via an integrated language model, and rendered by a randomly selected 3D avatar using the Heygen Streaming Avatar SDK. All interactions and translated phrases are logged and can be saved to a personal vocabulary list.

One of the main challenges was integrating multiple APIs (speech recognition, translation, avatar streaming, and data storage) into a seamless and user-friendly interface. We also had to build conditional logic to control which output language is allowed, based on the student’s language level. Managing microphone permissions across devices and ensuring low-latency performance in real time proved technically demanding.

We successfully created an intelligent, culturally adaptive tutoring tool that bridges the gap for multilingual students in secondary education. The system personalizes every aspect of the learning experience—from language direction to avatar identity—and stores relevant progress in a structured and accessible way. The end result is both technically robust and pedagogically effective.

Designing AI for education requires more than just technical proficiency; it demands empathy, adaptability, and a deep understanding of the learner’s context. We learned how to align technology with real classroom needs, and how meaningful personalization can boost both confidence and engagement in language learners.

We plan to extend the platform with features such as automated test preparation, teacher dashboards, and AI-generated exercises based on each student’s vocabulary list. We also aim to enhance avatar realism and add pronunciation feedback using AI, further strengthening the tutor’s role as a supportive, interactive learning companion.

Built With

• airtable
• airtable-(user-database-&-vocabulary-storage)
• apis
• bolt.new-(frontend-builder)
• heygen-streaming-avatar-sdk-(3d-avatar-output)
• html/css-(via-bolt/tailwind)
• javascript-(client-side-logic)
• openai-gpt-api-(translation-logic)
• openai-whisper-api-(speech-to-text)
• supabase-(authentication-&-hosting)