Work 02 · Human-centered AI
Designing accessible AI for hard-of-hearing drivers
The model was one part of the product. The harder question was how to translate critical sounds into useful alerts without adding distraction.
The human need
Environmental sounds carry safety information, but not every sound deserves equal attention.
I led a 0-to-1 mobile product for hard-of-hearing drivers, from problem framing and discovery through MVP delivery. The core product question was how to make emergency sounds timely and glanceable without creating another source of distraction.
Product and model decisions
We designed the experience and the classification pipeline as one system.
The MVP combined visual and haptic alerts with an audio-classification pipeline using YAMNet/AudioSet and an RNN. I connected React Native, Python, REST, and TensorFlow Lite work around the same response-time and usability constraints.
User feedback and field testing shaped the alert hierarchy and backlog rather than being treated as a final validation step.
Evidence and boundaries
The system reached 92% accuracy in field testing and produced a functional mobile MVP.
That result supports the project evaluation setup; it does not claim clinical effectiveness or proven road-safety impact. Future work should evaluate alert-specific false positives, latency, sustained use, and representative user experience.
Reflection
In high-stakes AI, usefulness, reliability, attention, oversight, and recourse have to be designed together.
- What I would do differently
- I would involve a broader group of representative users earlier and define the cost of each false-alert category before expanding the taxonomy.
- Framework connection
- Product Diagnosis Canvas · AI Trust Framework
Evidence boundary
Back to all transformations This case describes an academic project and its field-testing result. It does not claim clinical deployment or validated road-safety outcomes.