Back to Portfolio
Mobile & Machine Learning Pokhara City Transit

Smart Bus Arrival Detector

A commuter-first mobile app designed for tracking municipal buses in real time with machine-learning powered ETA predictions across Pokhara city routes.

Smart Bus Arrival Detector Interface Preview

Role

Full-Stack Mobile & ML Developer

Platform

Android & iOS (React Native)

Tech Stack

React Native Firebase Google Maps API Python (ML)

Impact / Result

Real-Time GPS Tracking • Route ETAs

The Problem & Background

In cities like Pokhara, public transit commuters face significant uncertainty regarding bus arrival times. Without a centralized dispatch or telemetry system, passengers frequently wait at bus stops for unpredictable intervals, and traffic bottlenecks fluctuate dynamically during peak hours.

The Smart Bus Arrival Detector was built to eliminate this commuter friction by broadcasting GPS telemetry from active transit buses and applying route-aware machine learning models to calculate dynamic arrival windows.

Key Features

Live Bus Radar

Visual map interface rendering moving bus coordinates with smooth marker interpolation and route waypoints.

ML ETA Forecasting

Dynamic arrival time computation factoring in historical road segment speeds, time-of-day traffic, and current velocity.

Stop Proximity Alerts

Configurable push notifications that alert commuters 5 and 2 minutes before the bus arrives at their selected stop.

Route Planning

Browse municipal transit routes, stops, and schedules with offline caching for frequent daily commutes.

Architecture & Data Pipeline

The system consists of three interconnected layers:

1. Telemetry Ingestion

Low-latency telemetry streaming using Firebase Realtime Database for lightweight, real-time push synchronization across mobile clients.

2. Prediction Engine

Regression models trained on route transit segment datasets calculate dynamic arrival windows instead of static distance-over-speed estimates.

3. Client Application

Cross-platform React Native frontend optimized for battery efficiency, minimizing GPS background polling while retaining high accuracy.

Key Engineering Challenges & Solutions

Challenge 1: GPS Drift & Map Snapping

In hilly terrain and narrow streets around Pokhara, raw GPS readings exhibited occasional jitter and drift away from the mapped roads. To solve this, a Kalman filter and road-network snap-to-polyline algorithm were applied to guarantee seamless marker trajectories on Google Maps.

Challenge 2: Battery & Data Bandwidth Consumption

Constant real-time network subscriptions on budget smartphones drain battery rapidly. We implemented adaptive polling intervals: when a bus is idle or distant, updates stream every 10 seconds, scaling up to sub-second updates only when approaching a designated commuter radius.

Portfolio All Projects Next Case Study Gate Automation