Understand & communicate
Camera-based recognition, contextual menu understanding and translation help turn unfamiliar information into a usable answer.
AI Travel Assistant · Independent Developer
TripPup
A published iOS travel assistant, independently built with SwiftUI, Next.js and Supabase. Task-specific tools turn travel questions into structured, actionable workflows.
The shipped product
Original product UI screenshots. Interface text shown in the images is Chinese.
Travel tasks with concrete outcomes
Menu understanding, itinerary generation and rescheduling, ticket parsing, nearby discovery and attraction guides address the need for contextual help during a trip.
Context, retrieval and tool execution
Implemented on-demand document retrieval, screen-context injection, structured context, OCR and function-calling workflows. The model can answer directly, retrieve information or invoke tools according to the task.
Task-specific tools
Plan & discover
Structured itineraries, rescheduling, nearby discovery and attraction guides connect natural-language requests with travel workflows.
Engineering decisions
Structured outputs support editable itinerary UI. Document access is scoped by sensitivity and user choice. Dedicated task entry points reduce the need for users to construct prompts.
Reliability and delivery
SwiftUI client → Next.js backend → Supabase authentication, PostgreSQL and storage. Implemented dual-model fallback, two-layer rate limiting, version compatibility, error recovery and anonymous trials. Independently handled App Store review and issue resolution.
My role
Independent developer responsible for product requirements, AI workflow design, the SwiftUI client, Next.js backend, Supabase integration, reliability features and App Store delivery.