Abstract
Text-to-SQL has gained increasing attention with Large Language Models (LLMs). While existing architectures have demonstrated the potential of multi-agent systems there remains significant room for improvement. In this work, we extend the CHESS framework by integrating In-Context Learning (ICL) techniques into the Candidate Generator module, evaluating three strategies: Zero-Shot, Few-Shot Learning, and Retrieval-Augmented Generation (RAG). We implement the system using GPT-4o, and perform experiments on the financial dataset from BIRD-SQL. Results show that Few-Shot Learning and RAG significantly outperform the standard approach. Compared to Zero-Shot (59.31% Execution Accuracy (EX), 0.412 ROUGE-1), RAG significantly boosted performance, increasing EX to 69.48% and ROUGE-1 to 0.652.

Abstract

Abstract
Purpose: The aim of the current study is to describe the functionality of free software developed for energy system contributions and energy expenditure calculation during exercise, namely GEDAE-LaB. Methods: Eleven participants performed the following tests: 1) a maximal cycling incremental test to measure the ventilatory threshold and maximal oxygen uptake (VO2max); 2) a cycling workload constant test at moderate domain (90% ventilatory threshold); 3) a cycling workload constant test at severe domain (110%VO2max). Oxygen uptake and plasma lactate were measured during the tests. The contributions of the aerobic (AMET), anaerobic lactic (LAMET), and anaerobic alactic (ALMET) systems were calculated based on the oxygen uptake during exercise, the oxygen energy equivalents provided by lactate accumulation, and the fast component of excess post-exercise oxygen consumption, respectively. In order to assess the intra-investigator variation, four different investigators performed the analyses independently using GEDAE-LaB. A direct comparison with commercial software was also provided. Results: All subjects completed 10 min of exercise at moderate domain, while the time to exhaustion at severe domain was 144 ± 65 s. The AMET, LAMET, and ALMET contributions during moderate domain were about 93, 2, and 5%, respectively. The AMET, LAMET, and ALMET contributions during severe domain were about 66, 21, and 13%, respectively. No statistical differences were found between the energy system contributions and energy expenditure obtained by GEDAE-LaB and commercial software for both moderate and severe domains (P > 0.05). The ICC revealed that these estimates were highly reliable among the four investigators for both moderate and severe domains (all ICC = 0.94). Conclusion: These findings suggest that GEDAE-LaB is a free software easily comprehended by users minimally familiarized with adopted procedures for calculations of energetic profile using oxygen uptake and lactate accumulation during exercise. By providing availability of the software and its source code we hope to facilitate future related research.

Abstract

Abstract
AbstractThe research on software startups has increased lately, focusing on describing how these companies’ unique context influences development practices. The next step for research is the creation of specific practices for these companies grounded in scientific results. An obstacle in this path is which dependent variable these novel practices should improve. A natural answer is these companies’ success. This position paper reviews the literature on new ventures and startups’ success to show that telling if a startup is successful or not is a complex issue. As a solution to this problem, this paper proposes a conceptual framework, suggesting that novel practices should improve success determinants or reduce inhibitors rather than focusing on the startups’ success. Three examples illustrate the framework’s use: hypotheses engineering, microservices, and BizDev. The identification of contributors and inhibitors for success of software startups could enrich the framework and indicate possible avenues for the creation of development practices specific tailored for these companies.