Veloquo
Turn English videos on iPhone or iPad into a dual-subtitle learning experience. Subtitle generation, translation, and explanations all run on-device.
Veloquo
Overview
Veloquo turns English videos stored on iPhone or iPad into a learning experience. Built on Apple's Speech, Translation, and Foundation Models frameworks, it provides dual subtitles, single-sentence loop playback, and tap-to-look-up dictionary entirely on-device. Videos never leave the device; only lightweight data such as subtitles and vocabulary is synced to the cloud.
The problem
Plenty of learners want to study with English videos, but cloud-based subtitle generation feels slow and raises privacy questions. The basics learners actually need — repeating one sentence, looking up a word in place — are missing from most video apps.
Key features
- On-device English subtitles paired with Japanese translation
- Streaming generation keeps subtitles in step with playback
- Single-sentence loop for repeated practice
- Tap a word in the subtitle to look it up in place
- Playback controls tuned for studying — 0.5–2× speed, 10s skip
Video & channels
Short videos start as on-device screen recordings from a real iPhone, then captions and SE are added in CapCut. The same source goes to YouTube Shorts, TikTok, and Instagram Reels — a 30-second entry point that shows the product in motion.
Same video also on 
TikTok .
Tech stack
Architecture
Built with SwiftUI and MVVM. Subtitles use the Speech framework, translation uses the Translation framework, and sentence-level explanations use Foundation Models. Videos and their intermediate data never leave the device; only lightweight data — subscription state, learning history — is stored in the cloud.
AI's role and boundaries
AI handles the sentence-level explanations, summaries, and SVOC breakdowns, with Apple Intelligence's Foundation Models keeping inference entirely on device. Subtitle generation and translation use the Speech and Translation frameworks, so removing inference would remove the core learning experience. What stays with the user is the choice of which sentence to loop and which word to save — AI never decides the study scope. Skipping a cloud LLM keeps both inference cost and privacy risk low. Quality is tuned using usage logs, with a focus on subtitle responsiveness and perceived latency.
Evaluation & Operations
Released on the App Store with a three-tier model — Free, Basic, Pro. We watch usage logs for loop frequency and perceived subtitle latency, then iterate on subtitle responsiveness and dictionary accuracy.