Check it Out
awards
WiWo Award TRUSTEQ
BrandEins Berater 2026 Award
TRUSTEQ kununu awards
TRUSTEQ
CooperativeExcellence

Custom SQL Agent with Claude 4.5 Sonnet: Intuitive natural-language queries of Claude resources – transparency regarding ownership, security, and costs.

Database Queries Reimagined: An LLM-Powered Solution for Intuitive Natural-Language Queries

As part of a project in the financial services sector, the goal was to significantly simplify access to complex databases for business units. The customer databases contained business-critical information about cloud resources, including, for example, VPC information, ownership data, security status, configuration parameters, and cost information for individual resources. Until now, these data queries were heavily dependent on technical expertise and SQL knowledge, which slowed down processes and tied up internal resources. Using a LangChain-based SQL agent in combination with the powerful Anthropic models Claude 4.5 Sonnet and Claude 3 Haiku, a solution was developed that precisely translates natural language into SQL queries. At the same time, the solution provides significantly greater transparency into the existing cloud landscape, as information on, for example, ownership, security status, and resource costs is available and traceable at any time. The accuracy of the SQL queries was approximately 97% in testing.
Schedule an expert consultation

Key facts about the AI-powered database solution

Practice of Case Study: AI for Process Optimization

Industry: Financial Services

Contact: Dr. Lukas Bonenberger (AI Managing Consultant), lukas.bonenberger@trusteq.de

The Challenge: Current Situation and Objectives

The goal was to create an application that allows employees to access relevant cloud data directly and transparently using simple voice commands.

Initially, the customer was working with several business-critical databases that stored information about cloud resources. This included VPC data, owner assignments, security statuses, tags, configuration parameters, and cost information for individual resources. Until now, using this data has been heavily dependent on technical expertise. Business departments needed quick access to information such as VPC owners, the security status of cloud resources, or cost overviews, but to do so, they either had to involve technical teams or possess SQL skills themselves. The solution had to overcome several technical hurdles:

  1. Converting natural language into correct SQL queries
  2. Integrating multiple databases on a shared server
  3. The ability to distinguish between database-relevant and irrelevant queries (intent detection)
  4. Transparent display of the generated SQL queries


The use of Claude 4.5 Sonnet, a powerful Anthropic model, was a key success factor, particularly for the highly precise query logic.

Relevance

The solution was essential for business agility, cloud governance, and process optimization, as previous database queries had led to inefficient processes and were heavily dependent on individual employees’ knowledge of SQL.

Project Objectives

Increased efficiency when interacting with databases through intuitive voice input. Improved transparency and clarity of the generated SQL queries. Optimization through cost savings and reduced technical complexity.

Our structured TRUSTEQ approach for measurable results

The project was carried out in two closely interlinked phases. Prototyping and production-ready implementation served as effective methods and tools. TRUSTEQ not only provided consulting services but also handled the technical design, development, and implementation of the solution, as well as employee training and enablement. What makes the solution unique is that it offers maximum transparency by translating generated SQL queries back into natural language, ensuring that results are presented in a clear and understandable manner. It also efficiently optimizes resources through the targeted combination of the powerful Anthropic models Claude 3 Haiku and Claude 4.5 Sonnet. These models impress with their high speed, excellent reasoning, precision, and cost-efficiency.

Analysis phase

Understanding of database structures, cloud resource models, user requirements, and business use cases related to the resources.

Prototyping and Implementation

Development of a LangChain-based SQL agent and implementation of the Anthropic models Claude 4.5 Sonnet and Claude 3 Haiku. Testing and optimization were conducted continuously in parallel with development to iteratively ensure precision, robustness, and user-friendliness. In particular, the continuous fine-tuning of the prompting and routing logic surrounding Claude 4.5 Sonnet was a key component of the process.

From concept to finished solution: Our approach

Our response to this challenge was to develop an intelligent system based on modern AI technologies. The solution combines the power of the LangChain SQL Agent with specialized Anthropic large language models. Employees can now, for example, ask questions in natural language such as “Which VPCs belong to the Payments team and have increased monthly costs?”, “Show me all resources without a defined owner,” or “Which resources are in a critical security status?” Claude 3 Haiku handles fast and efficient intent detection and checks whether the query is relevant to the database. Claude 4.5 Sonnet was used for high-precision SQL generation and the clear translation back into natural language.

The model stood out in particular for its high precision with complex SQL statements, strong contextual understanding across multiple data sources, and a highly reliable logical structuring of query dependencies. Additionally, a multi-database functionality was implemented, enabling seamless aggregation of multiple data sources within a unified interface. Claude 4.5 Sonnet played a central role in the robust processing of complex resource structures.

Furthermore, an AI explainability component was implemented, which transparently displays both the generated results and the underlying database query.

New components through successful implementation:

With new AI modules for natural language query processing and SQL validity, the back-translation of generated queries to enhance acceptance and traceability, and optimized resource management through the combination of specialized LLMs (Claude 4.5 Sonnet and Claude 3 Haiku), the solution delivers maximum efficiency and transparency.


Results: An Overview of the Benefits of Our AI-Powered Solution

The solution we developed achieved remarkable results and provided the client with direct benefits.

High precision

The SQL queries generated by Claude 4.5 Sonnet were correct in approximately 97% of cases.

User-friendliness and transparency

Thanks to the clear back-translation provided by Claude 4.5 Sonnet, more employees were able to use the application intuitively, and the transparency of the results was significantly improved.

Efficient use of resources

Thanks to the combined use of Claude 3 Haiku and Claude 4.5 Sonnet, the solution could be operated with reduced computational overhead.

Increased productivity

The elimination of manual SQL queries and reduced training requirements led to a significant reduction in the workload for the business units. At the same time, transparency regarding resources was significantly improved.

An overview of our solution’s key achievements

The solution boosts productivity and streamlines technical processes by providing intuitive access to databases and cloud resource information. Employees can focus more on their core tasks and gain faster visibility into their resource landscape.

Particularly noteworthy is the effectiveness of the combination of specialized Anthropic models: Claude 4.5 Sonnet impressed with its excellent precision, strong reasoning, and high reliability in SQL generation and explainability, while Claude 3 Haiku stood out for its high speed and efficiency in intent detection.

Dr. Lukas Bonenberger

AI Managing Consultant