Published September 2, 2026 | Version v2
Dataset Open

Dataset on System Performance, Usability, and User Acceptance of Edge-Based Self-Reflective RAG Health Assistant

Authors/Creators

  • 1. ROR icon Binus University
  • 2. ROR icon Universitas Dian Nuswantoro
  • 3. Semarang City Health Office
  • 4. ROR icon Diponegoro University

Description

This repository contains the empirical evaluation dataset and supplemental codebase for the research paper entitled "Dataset on System Performance, Usability, and User Acceptance of Emi: An Edge-Based Self-Reflective RAG Health Assistant" (submitted to Data in Brief, Elsevier).

1. Overview

Emi is an edge-AI conversational agent tailored for Psychological First Aid (PFA) and Non-Communicable Disease (NCD) public health education in Semarang City. The system runs entirely on consumer-grade edge hardware (≤8 GB VRAM) utilizing quantized Gemma 4, Qdrant vector database, and LangGraph orchestration.

2. Dataset Contents

  • 1_Raw_Data_N20.csv: Anonymized survey responses (N=20) evaluating Technology Acceptance Model (TAM) constructs and System Usability Scale (SUS) scores.
  • 2_Outer_Stage1_Comparison.csv: Measurement model construct reliability, Cronbach's Alpha, Composite Reliability, and Average Variance Extracted (AVE).
  • 3_Master_Outer_Loadings.csv: Cross-loadings and indicator factor matrix.
  • 4_Stage2_HOC_Weights_Loadings.csv: Formative and reflective Higher-Order Construct (HOC) parameters.
  • 5_Hypothesis_Bootstrapping.csv: Structural path coefficients, t-statistics, p-values, and mediation analysis results.
  • 6_DSR_Technical_Logs_N115.csv: System latency logs, Time-To-First-Token (TTFT), token throughput (TPS), and RAGAS accuracy evaluations across 115 benchmark queries.
  • 7_BlackBox_Safety_Test.csv: Boundary safety test cases, jailbreak resilience verification, and safety adherence evaluations.
  • 8_Survey_translated.docx: Translated Post-Interaction Questionaire for data collection

3. Supplemental Codebase

  • emi_core.py: Core LangGraph agent and prompt orchestration pipeline.
  • emi_api.py: FastAPI backend service handling decoupled streaming and PostgreSQL logging.
  • streamlit_emi.py: Frontend user interface with audio synthesis.
  • mem_fill.ipynb: Ingestion pipeline utilizing GOT-OCR 2.0 and Qdrant embedding injection.
  • docker-compose.yml: Container orchestration for PostgreSQL and Qdrant services.

Files

Emi_Research_Dataset.zip

Files (79.7 kB)

Name Size
md5:882a4c5ea81c19aaab9e11dc3a65e6b5
69.6 kB Preview Download
md5:e85f66a2c354abc361caa08e28902798
10.1 kB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/Atanb-code/Emi/tree/main (URL)

Software

Repository URL
https://github.com/Atanb-code/Emi/tree/main
Programming language
Python , Jupyter Notebook
Development Status
Active

References

  • Dinas Kesehatan Kota Semarang. (2024). Profil Kesehatan Kota Semarang Tahun 2024. Pemerintah Kota Semarang.
  • StepFun. (2024). General OCR Theory (GOT-OCR 2.0). StepFun AI.
  • Shah, C., et al. (2024). Ragas: Automated Evaluation of Retrieval Augmented Generation. arXiv preprint arXiv:2309.15217.
  • Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319-340.
  • Brooke, J. (1996). SUS: A 'quick and dirty' usability scale. Usability Evaluation in Industry, 189-194.