SIGAP — AI-Powered Urban Activity Decision Engine

Competition entry for #JuaraVibeCoding, a Google-organised developer competition involving Gemini API, Google Cloud Run, and Google Maps Platform.

Outdoor safety decisions in Indonesian cities often require checking multiple disconnected sources. SIGAP addresses this by combining environmental and mobility signals into one activity- and location-aware recommendation.

• Combine real-time environmental and mobility data into a unified decision.
• Use Gemini as a reasoning layer to produce structured, actionable recommendations.
• Deliver the result through a simple decision-focused web interface.

Data from multiple external services is collected and normalised server-side before being passed to Gemini as structured environmental context. The system is designed to gracefully handle individual provider failures while the frontend focuses on presenting the resulting recommendation rather than exposing raw data alone.

BMKG, Google Air Quality, Google Routes, PetaBencana, Visual Crossing, and Stormglass.

A Next.js application orchestrates external data sources through a server-side API layer, passes the resulting context to Gemini, and returns a structured recommendation to the client. The application is containerised with Docker and deployed on Google Cloud Run.

• Designing Gemini prompts that behave consistently across different interaction modes.
• Handling inconsistent external API behaviour, including PetaBencana's undocumented endpoint and lack of radius filtering.
• Resolving production deployment issues involving Cloud Build and client-side API key configuration.

A fully deployed web application combining six real-time data sources with Gemini to provide contextual safety recommendations through both natural-language queries and structured trip planning.

• Using an AI model as a structured reasoning layer requires explicit guardrails for different user intents rather than relying on a single generic prompt.
• Multi-source applications need graceful degradation because external APIs cannot always be assumed to behave consistently.

• Desktop-only interface.
• No automated tests or monitoring.
• In-memory caching does not persist across Cloud Run instances.