At a glance:
Role: Consultant Design Engineer
Key Deliverables: Language switcher, native-style autocomplete, AI next-word prediction, ergonomics, interaction design
Stack: Swift, SwiftUI, KeyboardKit, OpenAI API
Background
Superwhisper's AI-powered dictation was popular with users, however, its custom keyboard did not meet the standards of Apple's native keyboard. Issues included the need to exit the keyboard to switch languages, an intrusive autocomplete feature, and typing latency, all of which posed a risk to user retention.
Objectives
The project initially focused on seamless language selection, user-controlled autocomplete, and achieving performance parity with native keyboards. The scope was later expanded to include native suggestions and predictions after their potential impact became evident.
Technical Highlights
Language Switcher: This feature utilizes
KeyboardContextandLocaleProviderto display installed locales and is hidden if only one locale is present.Autocomplete & Suggestions Bar: A
PredictionViewoverlay was added, which animates into view only when the text field has content.Next-Word Prediction: To adhere to App Store guidelines, network traffic was routed through the main app instead of the keyboard extension.
Ergonomic Layout: The record button was moved to the left, and the minimum tappable target size was increased to 44 points.
Developer Experience: The project included an upgrade to KeyboardKit 9.4, resulting in a 12% reduction in latency. Additionally,
swift-formatand GitHub Actions were implemented, and a refactoring of audio-buffer error handling led to zero crashes in TestFlight.
Results
The following data is based on anonymised, project-wide aggregates from SuperUltra's internal analytics.
"Typing is as smooth as Apple's keyboard, but with suggestions that actually get me."
— Private beta tester, June 2025
Retro
Scope can grow—if ROI is clear.
Animations earn trust. Thoughtful micro-interactions can effectively highlight new features.
Invest in tooling early. Continuous integration was successful in preventing 14 regressions before the release.
Ongoing
On-device machine learning for offline predictions.
Dynamic font-sizing for improved accessibility.
Ongoing user experience testing with Superwhisper's global community.
