Enterprise AI Analysis
LLM-Based Geospatial Assistant for WebGIS Public Service Applications
The automation of public services represents a key area of development at the national level, with the main goal of facilitating citizens' access to comprehensive, integrated and high-quality services in the shortest possible time. National strategies emphasize the need to integrate open geospatial data and artificial intelligence into information, transparency and decision-making processes. The evolution of artificial intelligence, particularly large language models (LLMs), has led to the development of virtual assistants capable of understanding user requirements and providing answers in natural, easy-to-understand language. This paper presents directions for the development and use of large-language-model-based virtual assistants, focusing on their ability to understand and interact with the geospatial domain through an LLM API. Geospatial modeling contributes significantly to the automation of public services, but access to this technology is often limited by technical expertise or dedicated software programs. The development of AI-based virtual assistants removes these barriers, facilitating access, reducing time and ensuring transparency and accuracy of information. The proposed approach is implemented using a commercial large language model API, integrated with domain-specific geospatial functions and authoritative spatial databases. This study highlights practical examples of virtual assistants capable of understanding the geospatial field and contributing to the optimization and automation of public services in the country. In addition, the paper presents comparative analyses, challenges encountered and potential directions for future research.
Executive Impact: Key Metrics
Leveraging LLMs and geospatial tools can significantly enhance public service delivery, bringing tangible benefits in accuracy, efficiency, and accessibility.
0
Base Model Function Selection Accuracy
0
Fine-tuned Clarification Accuracy
0
Max LLM Orchestration Time
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
LLM Limitations in Geospatial Tasks
Traditional large language models often struggle with direct access to verified, real-time geospatial databases and lack specific focus on geospatial reasoning. While progress has been made in code generation and semi-automatic map creation, these models frequently produce incomplete workflows, incorrect operators, or hallucinated functions without proper integration with authoritative data sources and domain-specific tools. This significantly reduces their practical utility in accuracy-critical applications like environmental monitoring or natural hazard analysis.
Function Calling for Geospatial Integration
Function calling allows LLMs to interact with external tools and APIs in a structured manner, translating natural language queries into executable spatial operations. This mechanism enables LLMs to access real-time, verified information from external databases and perform specialized operations beyond their intrinsic reasoning abilities. By using predefined input-output schemas, function calling reduces hallucinations and improves reproducibility, combining natural language interpretation with precise external computation for more accurate and contextually relevant geospatial results.
Impact of Fine-Tuning on Model Behavior
Fine-tuning adapts pre-trained foundation models to specific tasks or domains by continuing training on smaller, specialized datasets. While it can improve domain-specific behavior, such as function selection and parameter extraction, it also carries risks like overfitting and loss of generalization, especially with narrow datasets. In geospatial contexts, over-specialization can lead to models failing to handle diverse input formats or generalizing poorly to unseen tasks, highlighting the critical balance between dataset diversity and model capacity.
WebGIS Application & Architecture
The proposed WebGIS application integrates an AI assistant, an LLM with function-calling capabilities, and a set of geospatial functions linked to authoritative databases. Users interact via natural language, which the assistant translates into function calls. These functions are executed on the server side using verified spatial datasets, returning results in both textual and visual forms. This approach reduces complexity, improves accessibility, and supports transparency in public services by bridging natural language with specialized geospatial operations, eliminating the need for dedicated GIS software expertise.
Model Performance Comparison
The study compared three LLM configurations: a base model, a base model with function calling, and a fine-tuned model with function calling. The base model with function calling showed higher function selection accuracy (89.5%), reflecting its tendency to select functions even with incomplete inputs. In contrast, the fine-tuned model demonstrated significantly higher clarification request accuracy (84.62%), indicating an improved ability to identify missing information and request user input before execution, albeit with a reduced function selection accuracy (30.77%) due to a conservative decision strategy.
Limitations and Future Research
Key limitations include strong dependence on external geospatial data quality, reduced generalization from fine-tuning with limited datasets, and deficiencies in natural language geocoding and real-time temporal awareness. Future work will focus on expanding and diversifying fine-tuning datasets, integrating robust geocoding and time-awareness services, and exploring hybrid architectures. Evaluating the system in real institutional settings with end-users from public administration will provide insights into usability, trust, and long-term sustainability for production-ready geospatial assistants.
89.5%
Function Selection Accuracy with Base LLM Model (with Function Calling)
Enterprise Process Flow: Function Calling Workflow
User Query Parsing
→
Structured Function Call Generation
→
Execution by External System
→
Integration of Results
| Feature | Base Model (with Function Calling) | Fine-tuned Model (with Function Calling) |
|---|---|---|
| Function Selection Accuracy (FSA) |
|
|
| Clarification Request Accuracy (CRA) |
|
|
| Generalization |
|
|
| Reliability in Public Services |
|
|
Case Study: Automated Land Parcel Information Retrieval
Scenario: A public service user, without advanced GIS expertise, queries the system with "Show details for my land parcel with identifier 27521".
LLM-Based Assistant Action: The assistant interprets the natural language request and, through function calling, invokes the details_parcel(parcel_id) function. This function connects to authoritative geospatial databases to retrieve comprehensive information.
Result: The system returns both a textual report and a visual representation on an interactive map. The report includes the parcel's perimeter (340.25 meters), area (4630.01 square meters), number of buildings (1), and building area (210.9 square meters). The parcel geometry and associated attributes are generated using cloud-based geoprocessing tools, ensuring verifiable and accurate results, directly within the WebGIS interface. This eliminates the need for manual data lookup or specialized software.
Calculate Your Potential AI ROI
Estimate the annual savings and efficiency gains your organization could achieve by implementing LLM-based geospatial solutions.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A clear path to integrating LLM-based geospatial assistants into your public services.
Phase 1: Discovery & Strategy
Identify core public service workflows suitable for LLM-based geospatial automation. Define key performance indicators and gather requirements from stakeholders.
Phase 2: Geospatial Tool Integration
Integrate existing authoritative geospatial services and APIs (e.g., cadastral data, satellite imagery, coordinate conversion) with the LLM through a robust function-calling mechanism.
Phase 3: Model Adaptation & Training
Develop and curate domain-specific instruction-response datasets. Fine-tune the LLM for improved function selection accuracy, parameter extraction, and contextual understanding in public service scenarios.
Phase 4: WebGIS Deployment & Pilot
Deploy the LLM-powered assistant within a user-friendly WebGIS application. Conduct pilot programs with end-users to gather feedback and refine the system for optimal usability and performance.
Phase 5: Continuous Improvement & Scaling
Establish monitoring for model performance and data accuracy. Expand integration to more complex workflows and diverse geospatial datasets, ensuring long-term sustainability and scalability.
Ready to Transform Your Public Services with AI?
Book a personalized consultation to explore how LLM-based geospatial assistants can optimize efficiency and accuracy for your organization.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat