PEDAL v1.5.0: Intelligent Archiving and LLM-Assisted Evaluation System for Educational Research

The PEDAL Lab has released version 1.5.0, integrating the LLM-as-a-Judge evaluation framework, automated keyword extraction, and academic search engine optimization to achieve end-to-end intelligent management from lab work to digital archives. This version adopts a dual-track architecture to serve internal collaboration and external dissemination, introduces an intelligent evaluation engine, automated metadata management, and version control mechanisms, lowers the threshold for using AI tools, promotes the development of open science in educational research, and plans to expand multimodal analysis capabilities.

The field of educational research faces dual dilemmas: massive teaching experiment data, evaluation results, and research notes need systematic management, while traditional data analysis methods struggle to meet the needs of rapid iteration. The PEDAL Lab was born to address these issues, committed to building a complete solution from 'lab to digital archive'. Version 1.5.0 marks an important progress in the project's intelligence and automation.

PEDAL v1.5.0 adopts a dual-track architecture:

• Internal Research Track: Focuses on high-fidelity collection and real-time analysis of experimental data, supports automatic import of multiple formats, and uses JSON-LD to semantically describe metadata to ensure discoverability and reusability.
• External Publication Track: Focuses on academic dissemination and long-term preservation of research results, integrates standards from platforms like HighWire Press, automatically generates academically standardized digital archives, and supports DOI registration and persistent links. The dual-track design balances sensitive data protection and rapid dissemination of academic results.

Version 1.5.0 introduces the 'LLM-as-a-Judge' evaluation framework, using models like Google Gemini to build a prompt library for educational scenarios. It evaluates teaching materials, student assignments, and research outputs from multiple dimensions such as alignment with teaching objectives and cognitive complexity, generating overall scores and diagnostic feedback. The system tracks the consistency of prompt strategies through statistical calibration, optimizes evaluation quality self-adaptively, and improves correlation with human expert judgments.

The Auto-Key module automatically extracts keywords from research documents, maps them to educational standard classification systems like Bloom's Taxonomy and NGSS, and generates standardized metadata tags. These tags are encoded in JSON-LD format and embedded in public archive pages, helping search engine crawlers understand academic attributes and improve visibility and ranking in academic searches.

PEDAL v1.5.0 implements full-cycle version management of research artifacts, recording all modifications from the initial draft of experimental design to the final paper. The Schema Validator module performs consistency checks before data storage, marks non-compliant data, and requires corrections. It integrates Grid database and cloud storage, supporting distributed storage and fast retrieval of large-scale datasets to facilitate cross-project meta-analysis.

PEDAL v1.5.0 lowers the threshold for using AI tools, allowing researchers without programming backgrounds to access LLM analysis capabilities; standardized data management promotes research reproducibility and sharing, driving open science. Future plans include expanding multimodal input (classroom videos, audio) analysis capabilities and developing an intelligent prompt recommendation system that automatically selects evaluation frameworks based on research questions.