Casper is a Tagalog-first, AI-powered speech companion. It uses real-time phoneme detection to identify exactly where a child's motor planning fails. Using backward chaining (e.g., “lan” → “ralan” → “Paaralan”), it provides "micro-therapy" sessions of 3 to 5 minutes, turning a parent’s smartphone into a high-frequency training tool that aligns with local school curricula.

We utilized a spec-driven development approach to build a custom audio processing engine capable of recognizing Tagalog-specific phonemes. The backend logic manages the "scaffolded learning" algorithm, while the frontend was designed for low-friction use by parents and children, ensuring the tech stays out of the way of the learning.

The "Language Barrier" was our toughest hurdle; most AI speech models are English-centric and fail to recognize Filipino speech nuances. We had to bridge this gap by fine-tuning our detection to prioritize Tagalog phonemes and developing a dynamic algorithm that could handle the complexity of backward chaining in real-time.

We successfully localized speech therapy. By creating a Tagalog-first tool, we’ve provided a lifeline to the 60% of families outside Metro Manila who lack access to specialists. We turned a ₱1,500/hour luxury into a free, accessible resource that empowers children to find their own voice.

We learned that the "system" often fails because it isn't built for the minority. Our journey with Casper taught us that for children with apraxia, the intelligence is there—only the "blueprint" for speech is missing. AI can serve as that blueprint, providing an empathetic solution to a structural medical shortage.

Our focus is on partnering with certified Speech-Language Pathologists (SLPs) to refine our AI models and ensure they meet the highest clinical standards. Moving forward, we aim to collaborate with educational institutions to integrate Casper into the classroom, providing teachers with a progress dashboard and making it a standard supplementary tool for inclusive education across the Philippines.

Built With

Share this project:

Updates