Cisco Technology, Inc. has secured a major milestone in artificial intelligence with a newly patented system for database query generation. This innovation focuses on the patent titled Structured query language generation using large language models. The patent describes a dual model architecture designed to accurately generate and correct structured query language statements using reinforcement learning.

Optimizing Automated Query Generation

Company Name: Cisco Technology, Inc.

Patent Title: Structured query language generation using large language models

Patent Abstract: In one embodiment, a method herein comprises: inputting, by a device, an input prompt to a first large language model to generate an output; computing, by the device, a reward metric in part by using a solver to process the output; tuning, by the device and based on the reward metric, a second large language model configured to correct errors of the first large language model using reinforcement learning; and using, by the device, the second large language model to correct an error of the first large language model.

Cisco Technology, Inc. has earned the prestigious Swanson Reed patent of the month award for May 2026 within the Artificial Intelligence, Software, Cryptocurrency and the Cloud industry due to its revolutionary approach to automating database interactions. This invention represents a significant advancement in software engineering by addressing the inherent unreliability of natural language interfaces when generating structured database queries. By introducing a secondary model dedicated exclusively to error correction, the system greatly improves the dependability of automated code generation.

The technical sophistication of this dual model architecture sets a new benchmark for machine learning applications. Integrating a specialized solver to calculate a reward metric ensures that the correction process is grounded in logical verification rather than mere probabilistic guessing. This reinforcement learning loop allows the secondary large language model to dynamically adapt and correct syntax or logic flaws, eliminating the hallucination issues that frequently plague standard generative models.

Furthermore, this method bridges the gap between complex database administration and natural language input, unlocking higher efficiency for enterprises. Winning this accolade highlights how the framework optimizes computing resources and streamlines software development lifecycles. It represents an outstanding technological milestone that paves the way for secure, reliable, and fully automated data management systems worldwide.

United States Research and Development Tax Credit Eligibility

To qualify for the Research and Development tax credit in the United States, an innovation must satisfy the specific criteria established by the Internal Revenue Service. This invention directly aligns with the fundamental pillars of the four part test required for eligibility.

• Permitted Purpose: The project must aim to create a new or improved product or process that improves performance, reliability, quality, or durability. This system introduces an entirely new framework for reliable database code generation.
• Elimination of Uncertainty: The development must involve solving technical uncertainties regarding the capability, method, or appropriate design of the system. Finding the correct configuration to accurately score model outputs using a solver requires overcoming technical uncertainty.
• Process of Experimentation: The engineering team must evaluate alternatives through modeling, simulation, or systematic trial and error. Determining the optimal reward metrics for reinforcement learning necessitates rigorous experimental testing.
• Technological in Nature: The research must rely on principles of computer science, data science, or artificial intelligence engineering. This patent heavily utilizes machine learning algorithms, database theory, and automated logic verification.

Practical Applications Meeting Research and Development Tax Credit Criteria

1. Designing and testing specific mathematical solvers to accurately quantify the syntax correctness of generated database queries under varying architectural schemas.
2. Engineering and optimizing the reinforcement learning reward feedback loop to minimize latency during the live tuning of the secondary error correcting language model.
3. Simulating multi turn interaction loops between the primary and secondary models to evaluate the error correction success rate across complex nested database queries.

”