Multiple models. Mutual scrutiny. Domain-precise evaluation

Itay Matane

GitHub

LLM EvalSphere is a contamination-resistant LLM evaluation suite that leverages cooperative–adversarial self-evaluation among multiple language models. Each model in the ensemble generates domain-specific question–answer challenges, including open-ended reasoning tasks and multiple-choice items. All participating models address these challenges, and each serves as an independent judge, scoring responses for correctness, usefulness, and domain fidelity. 

Turning electronic datasheets into actionable knowledge 

Niv Saban

GitHub

SpecExtract is an LLM-based system that analyzes datasheets and technical manuals of electronic components to extract key characteristics and operational properties automatically. The system identifies specifications such as supply voltage ranges, maximum current ratings, power dissipation limits, pin configurations, timing parameters, and recommended operating conditions. For example, from a MOSFET datasheet, it can extract threshold voltage, Rds(on), gate charge, and thermal resistance. At the same time, from a sensor manual, it may retrieve accuracy, sampling rate, communication protocol, and calibration requirements. SpecExtract streamlines engineering workflows by transforming dense technical documents into structured, easy-to-use information. 

Fine-grained understanding of student feedback 

Guy Yogev

GitHub

EduAspectInsight is an aspect-based sentiment analysis system designed to process student course reviews and extract detailed evaluations across multiple dimensions of teaching and course organization. The model identifies and scores aspects such as instructional clarity, engagement, course content quality, workload balance, assessment fairness, and overall organization. By converting unstructured feedback into structured insights, EduAspectInsight helps instructors and academic programs pinpoint strengths, uncover areas for improvement, and track changes in student perceptions over time.