Telematics Query Retrieval Using LLM-Generated Database Requests
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Managing and retrieving telematics data from a fleet of vehicles is challenging due to the vast amount of data, time-consuming report reviews, and difficulty in identifying specific outliers, trends, and patterns, especially when dealing with multiple vehicles.
Innovation Solution
A system utilizing a large language model (LLM) generates executable queries based on natural language requests to retrieve specific telematics data without direct access to the databases, incorporating contextual prompts and example queries to efficiently respond to user inquiries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If comprehensive telematics data reports are retrieved for fleet management, then complete information about vehicle operations is obtained, but the time required to review the data and identify specific insights increases significantly
Solution Approach 1:
The system extracts and retrieves only the specific telematics data points that are relevant to the user's natural language question, rather than returning comprehensive reports. The query generation engine translates natural language questions into targeted SQL queries that extract only necessary data fields, reducing review time while maintaining information completeness for the specific inquiry.
Solution Approach 2:
The system segments the large volume of telematics data into specific, queryable units based on the user's natural language question. By breaking down comprehensive reports into targeted data extracts, the system allows users to access specific insights without reviewing entire fleet reports, thus reducing time loss while preserving necessary information.
2Loss of information
If comprehensive telematics data is downloaded for analysis, then complete vehicle operation information is available, but the data storage space required increases significantly
Solution Approach 1:
The system extracts only the specific telematics data points needed to answer the user's natural language question, rather than downloading entire datasets. The query engine translates questions into targeted SQL queries that retrieve minimal necessary data, reducing storage requirements while maintaining information completeness for the specific analysis need.
3Ease of operation
If natural language processing is implemented to enable users to query telematics data without coding knowledge, then ease of operation improves, but the complexity of the system increases due to integrating LLM and query generation components
Solution Approach 1:
The system introduces a natural language processing intermediary layer that translates user questions into executable SQL queries. This intermediary (comprising the LLM and query generation engine) handles the complexity of data retrieval logic, allowing users to interact with complex telematics data through simple natural language without needing to understand the underlying system complexity.
Solution Approach 2:
The system enables users to perform data retrieval themselves through natural language queries, with the query generation engine automatically translating their questions into appropriate SQL queries. This self-service approach empowers users to access needed information directly without requiring programming knowledge or assistance from data specialists, despite the underlying system complexity.
Data Source
AI summary
at least one data storage operable to store at least a plurality of databases, each database storing telematics data originating from a plurality of telematics devices installed in a plurality of vehicles; and at least one processor in communication with the at least one data storage, the at least one processor operable to: generate training data for training a machine learning model by: generating a natural language request comprising a textual question relating to the telematics data by inputting into the machine learning model at least: a contextual prompt providing to the machine learning model at least one or more features of the plurality of databases, and instructions to generate the natural language request based on the contextual prompt; generating an executable query for retrieving a portion of the telematics data that is responsive to the natural language request from one of the plurality of databases by inputting into the machine learning model at least the natural language request; executing the executable query; determining whether the executable query was successful in retrieving the portion of the telematics data; and generating at least a portion of the training data comprising the natural language request and the executable query; and input the at least portion of the training data into the machine learning model, thereby training the machine learning model.