EdgeCity

Inspiration

Thousands of students travel to international cities each year for academic pursuits. During the initial weeks, they struggle with unfamiliar languages and complex city transport systems. For example, a student from Singapore moving to Germany for a bachelor’s program faces immediate challenges navigating daily life. They may be unable to secure a local sim card, resulting in limited or no access to reliable internet services. EdgeCity was conceived to support users during this critical adjustment period with reliable, offline assistance.

What it does

EdgeCity provides offline navigation, translation, and decision-making in unfamiliar cities. It supports image-based sign and menu reading using external OCR with on-device translation. The app uses DeepSeek for reasoning over transport routes(based on pre-downloaded City Bundles) and Llama for dialogue and language translation. All functionality runs entirely on-device, without internet dependency.

How we designed it

EdgeCity is currently a conceptual project, developed as part of this ideation hackathon. It is designed using the RunAnywhere SDK for on-device model orchestration. Each City Bundle contains maps, transport networks, and language packs (~100MB per city). The inference pipeline follows: STT / OCR → RunAnywhere SDK → Llama / DeepSeek → TTS

Challenges we might run into

Balancing RAM usage and reasoning performance on mobile hardware. Might encounter some hurdles in designing compact city bundles that retain meaningful navigation and transport intelligence. Ensuring smooth model orchestration on lower-RAM devices.

Accomplishments that we're proud of

EdgeCity demonstrates that offline-first city intelligence is feasible on mobile devices. It enables instant, privacy-first, zero-latency AI guidance without cloud dependency. The concept showcases a practical use of on-device reasoning and multimodal AI.

What we learned

On-device AI requires careful model selection, quantization, and memory orchestration. The primary concept of Edgecity revolves around offline city guidance. Effective offline city guidance depends heavily on well-structured, localized data bundles. Privacy-preserving intelligence can coexist with real-world usability.

What's next for EdgeCity

Expand City Bundles to include more global cities and transport systems. Improve route accuracy and multimodal understanding (voice, image, and text). Explore personalized offline recommendations based on user preferences.

Built With

• none