Construction and Application of Intelligent Questioning Intelligent Agents Based on Large Models
Revolutionizing Enterprise Data Query with LLM-Powered Agents
Authors: Hongwei Wu, Shaolei Wang, Xinchen Wang, Hao He, Yuxiao Zhao, Liujian Diao
Abstract: The traditional development of query systems mainly relies on R&D personnel for coding and development, with a long development time cycle and difficulty in adapting to the dynamic and changing data query requirements in digital transformation as business develops. In order to enhance the flexibility and stability of the system, this article proposes to use artificial intelligence big model capabilities to build a question and data intelligent agent. By utilizing a knowledge base+Agent and low code orchestration, it can achieve functions such as user question completion, user question knowledge rewriting, user intent recognition, and tool calling. Users can perform data queries and knowledge Q&A through dialogue. After experimental verification, the intelligent question answering agent designed in this article can accurately answer user questions and greatly shorten the development cycle, meeting most of the current question answering and knowledge question answering needs.
Keywords: large Language Model, intelligent agents, low code orchestration
Published: 14 November 2025 | Downloads: 5 | Citations: 0 | Context: AIFM 2025, Guangzhou, China
Executive Impact & Key Metrics
Our analysis quantifies the immediate benefits for your enterprise by adopting LLM-powered intelligent agents.
0%
Development Cycle Shortened
0%
Query Accuracy for Relevant Queries
0
Total Downloads
0
Total Citations
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Core Competencies
System Architecture
Key Functionalities
Application Scenario
The proposed intelligent agents possess advanced core capabilities beyond traditional systems. These include:
- Multi-Round Rewriting: Utilizing large models to intelligently rewrite user questions, leveraging historical context for more accurate and relevant queries. For example, automatically completing dates and keywords in follow-up questions like "What's the weather like on December 12, 2025?" after an initial query for December 11th.
- Retrieval-Augmented Generation (RAG): Enhancing large models with local business knowledge bases to overcome limitations in specific scenarios. This allows the model to accurately rewrite keywords and understand domain-specific terminology, such as mapping "receivable electricity fees" to "sales amount" in the electricity marketing domain. This process involves document formatting, content extraction, and synchronization with an ES library.
- Summary Interaction: Employing prompt engineering to guide large models in summarizing output results according to user needs and rendering them in desired templates, significantly shortening development cycles for reporting and display pages.
The intelligent questioning agents are built upon large model capabilities, integrating various components to achieve their advanced functionalities. The architecture leverages the Alibaba Bailian platform, which provides access to multiple models including DeepSeek distillation (32B), DeepSeek inference (32B), DeepSeek dialogue (14B), Qianwen semantic (14B), and Qianwen 72B.
Key architectural components:
- Model Task Nodes: Facilitate the execution of various large model capabilities.
- Knowledge Retrieval Tools: Enable the system to access and utilize pre-configured business knowledge bases.
- Retrieval Enhancement Tools: Improve the accuracy and relevance of knowledge retrieval.
- Self-Translated Script Tools: Support custom logic and integration with existing systems.
- Low-Code Orchestration: Allows for flexible combination and management of different model capabilities and tools, empowering business personnel to configure agent behaviors without extensive coding.
The intelligent agents offer several key functionalities designed to improve data query and knowledge Q&A in enterprise settings:
- Scene Guided Intent Understanding: Rewrites user questions using historical context, extracts entities (units, time, indicators) guided by business knowledge, and classifies user intentions to accurately call sub-agents for data queries. The Qianwen Semantic Model-72B is used for entity extraction, and DeepSeek Dialogue Large Model-14B for intent understanding.
- Information Integration: Builds and utilizes a comprehensive business data knowledge base, including indicator meanings, data table structures, field meanings, and interface function explanations. This knowledge base, managed via the Alibaba Bailian platform, significantly enhances the large model's capabilities in knowledge rewriting, entity extraction, and intent recognition.
- Data Query Module: Supports single-dimensional, multi-dimensional indicator queries, and data table queries. It can integrate with original system interfaces by adding function descriptions to the knowledge base. For data table queries, the large model (DeepSeek Distillation Large Model-32B) understands user input, performs table/column recall, generates SQL statements, and returns encapsulated results.
The intelligent questioning agents have been successfully applied in the Jiangsu Province electric power marketing intelligent questioning scenario. This application addresses high-frequency and natural language data query demands from business personnel, who traditionally relied on IT teams for interface development.
The system provides core marketing intelligence functions:
- Asking for Data: Users can query specific data points and metrics (e.g., "Nanjing high-voltage user arrears details").
- Asking for Indicators: Retrieval and analysis of business indicators.
- Asking for Reports: Generation and retrieval of various business reports.
- Other Functions: Such as electricity fee file management.
Enterprise Process Flow: Intelligent Questioning Agent
User Query Input
→
Intelligent Understanding & Rewriting
→
System Validation & Tool Selection
→
Data Access & SQL Generation
→
Result Presentation
| Model Type | Typical Model | Core Competencies | Application Scenarios |
|---|---|---|---|
| Dialog Model | DeepSeek Dialogue Large Model-14B | Context understanding and multi round dialogue management | Problem completion, Intent recognition |
| Reasoning Model | DeepSeek Distillation Large Model-32B | Complex logical reasoning and numerical calculations | Analysis of indicator formulas and data correlation analysis |
| Semantic Model | Qianwen Semantic Model -14B (and 72B mentioned in text) | Entity recognition, Relationship extraction | Business terminology mapping and indicator caliber matching |
Case Study: Jiangsu Electric Power Marketing Application
In the Jiangsu Province electric power marketing scenario, the intelligent questioning agent has demonstrated significant value. Business personnel frequently need to query various indicators and data, which traditionally required IT team development, leading to low efficiency and limited adaptability. By leveraging large models and a dedicated business knowledge base, the system enables users to perform complex data queries and knowledge Q&A through natural language dialogue. It accurately identifies relevant data tables and columns, generates SQL, and presents results, fulfilling core marketing intelligence functions like asking for data, indicators, and reports. This application has greatly reduced development cycles and dependence on technical teams, while improving data utilization efficiency across the enterprise.
0%
Estimated Reduction in System Development Cycle
Calculate Your Potential ROI
Estimate the tangible savings and reclaimed productivity your organization could achieve with AI-powered intelligent agents.
Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A phased approach to integrate intelligent agents seamlessly into your enterprise operations.
Phase 1: Discovery & Strategy
Comprehensive assessment of your current data query workflows and business needs. Define clear objectives and a tailored strategy for intelligent agent deployment. Identify key knowledge sources and integration points.
Phase 2: Knowledge Base & Model Configuration
Build and integrate your enterprise-specific knowledge base (indicator meanings, data tables, interface descriptions). Configure and fine-tune large models for optimal performance in your domain. Implement low-code orchestration for agent logic.
Phase 3: Pilot Deployment & Iteration
Deploy intelligent agents in a controlled pilot environment. Gather user feedback and conduct iterative refinements to enhance accuracy, usability, and functionality. Ensure seamless integration with existing systems.
Phase 4: Full-Scale Rollout & Optimization
Expand agent deployment across relevant departments. Establish monitoring and continuous learning mechanisms for ongoing performance optimization. Provide training and support to maximize user adoption and impact.
Ready to Transform Your Enterprise?
Connect with our AI specialists to explore how intelligent questioning agents can drive efficiency and innovation in 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