Fleet Insights

Client Overview

A provider of fleet management and payment solutions sought to simplify fleet data analysis and identify cost reduction opportunities through AI-driven automation.

The Challenge

The client needed a way to automate fleet data analysis and deliver actionable insights to their customers. Manual data reviews were time-consuming, and existing tools lacked intelligent automation. Additionally, the AI model had to be scalable, ensuring it could process increasing amounts of data while maintaining accuracy and security. A key challenge was integrating Large Language Model (LLM) capabilities to generate reliable cost-saving recommendations while maintaining compliance with security protocols.

The Solution

York provided a structured AI-driven solution, leveraging LLMs, secure data handling, and architectural improvements to create an automated fleet data analysis system. Key Solution Components:

• Fleet Data Analysis Simplification:
  • Developed an AI model capable of analyzing fleet data and providing cost reduction insights through automatic prompting.
  • Integrated the AI-generated insights into a widget that could be embedded into multiple client-facing products.
• Collaborative Development & LLM Integration:
  • Developed an LLM-based solution that could answer over 40 fuel-related and 10 telematics-related questions.
  • Designed a prompt graph to enable automated responses.
• Architectural Improvements:
  • Split a monolithic codebase into five repositories for better project management and scalability.
  • Implemented Azure DevOps pipelines for seamless version releases.
• Secure Data Query Generation Architecture:
  • Designed a vector database-based approach to ensure that AI-generated queries were restricted to pre-approved data requests, preventing unauthorized access.
• Technology Stack & Platform Integration:
  • Used Docker to containerize the prompt service for deployment.
  • Integrated with the client’s authentication model to ensure secure user access.
  • Modified the Insights Prompt Service and Insights Prompt Graph to query the client’s Snowflake instance for data storage and retrieval.
• Knowledge Transfer & Best Practices:
  • Applied Test-Driven Development (TDD) methodologies to ensure high-quality outputs.
  • Conducted knowledge-sharing sessions to enhance internal development processes.

The Results & Impact

The AI-powered fleet data analysis system provided significant benefits, including:

• Reduced manual effort in fleet data analysis, saving time for end users.
• Automated cost-saving insights, enabling customers to make informed financial decisions.
• Scalable AI model, capable of handling growing data volumes without performance loss.
• Secure data query processing, ensuring compliance with strict data security requirements.
• Improved project scalability through repository restructuring and DevOps integration.
• Successful LLM implementation, achieving an accuracy score of over 90 points.