Spikdi

The problem

Language learners abandon apps that feel generic. A single fixed curriculum bores an advanced speaker and overwhelms a beginner, and most apps still cannot tell you whether you actually pronounced a word correctly. We set out to build a language app where lessons adapt to the learner and the app can hear and coach you — an AI product good enough to prove ISEMI ships production AI, not demos.

What we built

Spikdi is a mobile app that orchestrates three AI models behind one clean learning experience. It generates lessons tuned to each learner, listens to their pronunciation and coaches it in real time, and illustrates and narrates every lesson so the material is vivid instead of a wall of text. The result is a lesson that feels made for you, with a coach that actually listens.

• Gemini-powered lessons adapted to each learner's level and goals
• Whisper pronunciation coaching with instant spoken feedback
• Imagen visuals and natural AI narration in every lesson
• A mobile experience learners open every day

How it is built

The app is a React Native client over an orchestration layer that routes each learning moment to the right model: Gemini for lesson generation and adaptation, Whisper for speech recognition and pronunciation scoring, and Imagen for on-lesson visuals. Narration is generated automatically, so no lesson ships without audio. The models sit behind one product surface, so the learner never sees the machinery — they just get a lesson that fits and a coach that listens.

• React Native mobile client
• Gemini for lesson generation and adaptation
• Whisper for pronunciation recognition and coaching
• Imagen for lesson illustration, plus AI narration

The result

Spikdi runs in production at spikdi.com, serving real learners every day and standing as ISEMI's flagship proof that we can take a multi-model AI product from idea to a polished, shipping app. The same discipline — model orchestration, evaluation, and production operation — is exactly what we bring to client agent work.

• Live in production at spikdi.com, used by real learners
• Three AI models orchestrated behind one seamless mobile experience
• Lessons that adapt per learner, with spoken pronunciation feedback
• ISEMI's flagship proof of shipping multi-model AI to production