FleetGPT
FleetGPT is an agentic AI fleet-analytics solution adopted by Novus Hi-Tech to investigate fleet events and support driver-safety interventions.1 It was developed using AWS services and the Strands Agents SDK, addressing operational needs that traditional dashboards and manual event review could not handle at scale.1
Purpose and capabilities
Novus Hi-Tech operates thousands of vehicles across 10 countries, generating dashcam footage, alerts, GPS data, and speed warnings for review.1 FleetGPT gives operators a natural-language interface for querying fleet data, retrieving evidence, and searching policy documents.1 Its analyst layer supports conversational analytics and investigation, while its intervention engine monitors critical events and can initiate AI-powered voice calls to drivers with human approval.1
Architecture
The solution separates data and retrieval functions: Amazon Aurora PostgreSQL stores structured telemetry and supports Text-to-SQL, Amazon DynamoDB provides low-latency access to high-velocity event streams, and Amazon Bedrock Knowledge Bases indexes unstructured standard operating procedures, policies, and documentation.1 A supervisor agent routes work to specialized sub-agents; the Text-to-SQL agent plans, generates, and executes queries.1
Reported outcomes
Novus Hi-Tech reported a 46% reduction in road-behavior violations and a 55% decrease in drowsiness alerts per 1,000 km after adopting FleetGPT.1
“46 percent annual reductions in road-behavior violations.”
“55 percent reduction in drowsiness alerts per 1,000 km.”
FleetGPT demonstrates a workflow that connects telemetry analysis with evidence retrieval and human-approved intervention rather than limiting fleet analytics to dashboards.1
Related Pages
- Amazon DynamoDB — DynamoDB provides FleetGPT with operational records, vector search, and low-latency access to fleet events.
- Amazon Bedrock — Bedrock supplies FleetGPT with managed foundation-model and agent capabilities, including retrieval over operational policy documents.