CS490 Project Topics

Ø Iterative and Multi-Path Query Enrichment with Large Language Models for Text-to-SQL

Yiğit Kemal Can

In this project, the first component, iterative query enrichment, progressively augments queries with database components (tables, columns, values) and candidate conditions, enabling models to resolve ambiguities and better connect with complex schemas. Complementing this, a multi-path enrichment strategy generates multiple enriched variants of a query in parallel, rather than recurrently, to explore diverse representations of user intent. These enriched queries will then be used for Text-to-SQL translation, and their impact on accuracy, schema linking quality, and error reduction will be systematically evaluated.

Ø A Unified Metric for Evaluating Accuracy and Efficiency in Schema Filtering for Text-to-SQL

Serhat Çelik

This project will evaluate a novel evaluation metric that jointly captures both accuracy (precision and recall of schema filtering) and efficiency (execution time and resource usage), providing a unified benchmark for Text-to-SQL systems. Existing evaluation protocols often treat correctness and speed separately, limiting the ability to assess trade-offs in real-world applications. By combining these dimensions into a single composite metric, this work will enable a more holistic comparison of schema filtering methods, particularly in scenarios where minimizing large language model calls is critical. The proposed metric will be validated on benchmark datasets and applied to advanced schema filtering strategies, offering a standardized tool for both academic studies and industrial deployments where performance and cost-effectiveness must be balanced.

Ø User-Centric Web and Desktop Interfaces for Text-to-SQL

Emir Görgülü

In this project, the Text-to-SQL methods developed by our research group will be transformed into practical, user-friendly applications by creating both web-based and desktop interfaces. The web interface will allow non-technical users to input database schemas, issue natural language queries, and receive SQL translations seamlessly through a modern, interactive platform. Complementing this, a desktop application will support offline use, ensuring data privacy for organizations that cannot rely on cloud-based services. Both interfaces will include features for schema management, query execution, and visualization of results, making advanced database access accessible to a broader range of users. By bridging cutting-edge research with real-world usability, the project will extend the impact of Text-to-SQL technology from academic development to practical adoption.