datablogs

Smart Buses, Smarter Operations: Building an AI-Powered Fleet Platform on AWS

Selva September 05, 2026 0 Comments

Introduction

Public transportation systems generate massive volumes of operational data every second. Bus locations, passenger occupancy, delays, fuel levels, traffic conditions, and breakdown events all contribute valuable insights that can improve service quality and operational efficiency.

In this proof of concept (POC), we built a complete cloud-native Smart Bus Operations Platform capable of:

  • Simulating real-time bus telemetry
  • Streaming IoT data into AWS
  • Storing historical operational data in a data lake
  • Automatically discovering schemas
  • Leveraging Generative AI to answer fleet-related questions
  • Producing operational summaries and recommendations

The solution was implemented entirely using AWS managed services with minimal infrastructure management.

Business Problem

Transportation operators often struggle with:

  • Lack of real-time fleet visibility
  • Delayed incident detection
  • Manual operational reporting
  • Poor route optimization
  • Limited predictive insights
  • Difficulty answering operational questions quickly

Examples include:

  • Which buses are currently delayed?
  • Which routes are affected by traffic?
  • Which buses require refuelling?
  • What is the overall fleet health?
  • Which vehicles need immediate attention?

Traditionally, answering these questions requires multiple dashboards, reports, and manual analysis.

Generative AI combined with IoT telemetry can significantly simplify operations.

Solution Architecture

End-to-End Architecture

Architecture Components

1. Bus Telemetry Simulator

A Python-based simulator was developed to emulate real-world bus operations.

The simulator generated:

  • Bus ID
  • Route Information
  • GPS Coordinates
  • Speed
  • Fuel Level
  • Passenger Count
  • Occupancy
  • Delay Information
  • Traffic Events
  • Breakdown Events
  • Weather Impacts

Sample Event

{

  "busId": "TN38-1020",

  "route": "Coimbatore-Palakkad",

  "speed": 49,

  "occupancy": 41,

  "fuelLevel": 27,

  "delayMinutes": 4,

  "event": "NORMAL"

}

The simulator continuously streamed data for multiple buses simultaneously.

2. AWS IoT Core

AWS IoT Core acted as the ingestion layer.

Responsibilities:

  • Device Authentication
  • MQTT Communication
  • Secure Message Transport
  • Rule Processing

Telemetry was published to:

bustime/telemetry

Each simulated bus behaved like a real IoT device.

 

 

 

3. Amazon S3 Data Lake

All telemetry data was persisted into Amazon S3.

Benefits:

  • Durable storage
  • Low cost
  • Unlimited scalability
  • Historical data retention

Storage Structure:

s3://bustime-iot-data/

 

raw/

 ├── event1.json

 ├── event2.json

 ├── event3.json

This became the system of record for all operational events.

 

4. AWS Glue Data Catalog

AWS Glue automatically discovered the telemetry schema.

Benefits:

  • No manual schema creation
  • Automatic metadata management
  • Simplified data discovery

Detected fields included:

  • busId
  • route
  • speed
  • occupancy
  • fuelLevel
  • delayMinutes
  • event
  • timestamp

This transformed raw IoT data into a searchable dataset.

 

 

5. Amazon SageMaker Studio

Amazon SageMaker Studio served as the analytics and AI development environment.

Responsibilities:

  • Data exploration
  • Fleet analytics
  • AI integration
  • Operational reporting

Python notebooks were used to:

  • Load telemetry data
  • Create fleet KPIs
  • Generate route summaries
  • Prepare AI prompts

 

6. Amazon Bedrock

Amazon Bedrock enabled natural language interaction with operational data.

Model Used:

Amazon Nova Micro

Benefits:

  • Lowest inference cost
  • Fast response times
  • Ideal for operational assistants
  • No infrastructure management

 

 

Fleet Intelligence Layer

Before sending information to Bedrock, fleet metrics were calculated.

Examples:

  • Total Active Buses
  • Delayed Buses
  • Breakdown Count
  • Low Fuel Vehicles
  • Average Route Delay
  • Route Performance Metrics

These metrics were included in prompts sent to Bedrock.

 

AI-Powered Operations Assistant

Users can ask natural language questions such as:

Fleet Health

Summarize fleet health

Delays

Which routes are experiencing delays?

Maintenance

Which buses require immediate attention?

Fuel Monitoring

Which buses need refueling?

Executive Reporting

Generate a fleet operations report.

Sample AI Response

 

 

Why Bedrock Instead of Traditional ML?

Traditional ML approaches often require:

  • Model training
  • Feature engineering
  • Infrastructure management
  • MLOps pipelines

Bedrock enables:

  • Zero model training
  • Immediate deployment
  • Natural language interaction
  • Faster POC execution

This significantly reduces implementation effort.

Alternative Architecture 1: Real-Time Operational Assistant (Recommended for Production)

 

Benefits:

  • Real-time bus status
  • Millisecond lookups
  • Scalable architecture
  • Conversational assistant

Example Questions:

  • Where is Bus TN38-1020?
  • Which buses are delayed right now?
  • Show active breakdowns.

Alternative Architecture 2: Analytics-Centric Architecture

Benefits:

  • Historical analytics
  • Trend analysis
  • Route performance insights
  • Executive reporting

Ideal for management dashboards and reporting.

Alternative Architecture 3: Knowledge Base Driven Assistant

Suitable for:

  • Bus schedules
  • Route maps
  • Operating procedures
  • Service manuals
  • FAQ systems

Best for static operational knowledge rather than live telemetry.

Benefits Achieved

Operational Visibility

Real-time fleet awareness.

Faster Incident Response

Immediate identification of breakdowns and delays.

AI-Powered Insights

Natural language access to operational data.

Reduced Costs

Serverless architecture with pay-as-you-go pricing.

Scalability

Capable of supporting thousands of buses.

Future Enhancements

Potential future enhancements include:

  • Real-time ETA prediction
  • Predictive maintenance
  • Driver behavior analysis
  • Passenger demand forecasting
  • Route optimization
  • Geospatial visualization
  • Fleet digital twin implementation

Conclusion

This POC demonstrates how AWS managed services can be combined to rapidly build an intelligent transportation platform.

Using AWS IoT Core, Amazon S3, AWS Glue, Amazon SageMaker Studio, and Amazon Bedrock, organizations can transform raw vehicle telemetry into actionable operational intelligence without building complex machine learning infrastructure.

The result is a scalable, cost-effective, AI-powered Smart Bus Operations Platform capable of improving fleet efficiency, enhancing passenger experiences, and enabling data-driven decision making.

0 Comments