
HacX 2025
This project aimed to revolutionize Singapore's NPPs by adding agentic AI features and designing a voice-first UI.
π HacX β Team Galileo
HACX 2025 Hackathon Project | Reimagining Singapore's Neighbourhood Police Posts
π― Problem
Singapore Police Force NPPs/NPCs handle many transactions that don't require direct officer involvement β yet the current self-help infrastructure is rigid, frustrating, and inaccessible:
- 9.8% of elderly citizens (65+) own 0 digital devices, yet they make up 20.7% of the population
- Existing self-help kiosks require users to manually fill forms in English, using complex police terminology
- The learning curve is very steep
- The current Found Property Drop Box has no monitoring capability, and fragile items are at risk of being broken
- Although an "Officer Clif" chatbot already exists, it is limited and not voice-driven
π‘ Our Solution
Reimagine NPPs/NPCs as friendly, intelligent, autonomous hubs.
We built a Voice-First, AI-powered self-help kiosk system that transforms how citizens interact with the police post β no typing, no jargon, no barriers.
Our system is:
- β NOT dictation
- β NOT text-to-speech
- β A human voice agent that is intelligent and adapts to the situation
β¨ Key Features
ποΈ Voice-First Incident Reporting Interface
Citizens can walk up and talk to the kiosk in natural language. The AI voice assistant β Officer Clif β guides them through the reporting process conversationally, in their own language.
- Powered by Azure AI VoiceLive (real-time WebSocket API) and GPT-4o Realtime Preview
- Dynamically fills in the report form as the user speaks
- Confirms each field back to the user before committing
- Supports multiple languages (responds in the user's spoken language)
- Professional, calm, and empathetic persona designed specifically for policing contexts
π¦ Smart Found Property Drop Box
A redesigned drop box that allows citizens to return lost items they have found.
- Modular design for needs-based construction with less complexity
- Hardware interface allows officers to monitor the contents of the box remotely
- Communicates with backend via serial (Arduino) + Flask API
- Reduces the need for fragile items to be handled unsafely
π₯οΈ Self-Help Kiosk β Incident Reporting
A clean, accessible web interface for the Woodleigh Neighbourhood Police Post.
- Multi-section guided forms with smart validation
- Voice assistant embedded directly into the reporting flow
- Handles both traffic incidents and general police reports
ποΈ Architecture
ββββββββββββββββββββββββββββββββββββββββ
β Next.js 15 Frontend β
β (React 19, TypeScript, Tailwind CSS)β
β β
β ββββββββββββββββββ βββββββββββββββ β
β β Report Form β β Voice Agent β β
β β (multi-step) β β (Officer β β
β β β β Clif) β β
β ββββββββββββββββββ ββββββββ¬βββββββ β
βββββββββββββββββββββββββββββββΌβββββββββ
β WebSocket (Azure VoiceLive)
ββββββββββββΌβββββββββββ
β Azure AI Services β
β GPT-4o Realtime β
βββββββββββββββββββββββ
ββββββββββββββββββββββββββββββββββββββββ
β Flask Python Backend β
β (Flask, Flask-CORS, PyAudio) β
β β
β - Serial interface to Arduino β
β - Drop Box LED / sensor control β
β - Azure Identity integration β
ββββββββββββββββββββββββββββββββββββββββ
β Serial (USB)
ββββββββββΌβββββββββ
β Arduino / IoT β
β Hardware Layer β
βββββββββββββββββββ
π οΈ Tech Stack
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, React 19, TypeScript, Tailwind CSS v4 |
| Voice AI | Azure AI VoiceLive, GPT-4o Realtime Preview |
| Backend | Python 3.13, Flask, Flask-CORS |
| Hardware | Arduino (Serial over USB via @serialport) |
| Cloud | Azure AI Services, Azure Identity |
| UI Components | Radix UI, Lucide React |
π Getting Started
π The Bigger Vision
The Future of Policing: Intelligent Agentic Models that can have conversations with users.
This project is a step toward making every Neighbourhood Police Post an accessible, intelligent, and autonomous hub β reducing officer workload on routine transactions while dramatically improving the citizen experience, especially for elderly and non-English-speaking residents.
Built at HacX 2025 by Team Galileo π