AI / ML

The REDI framework addresses the gap in unifying data transformation, readiness assessment, and provenance tracking for large-scale scientific datasets. It operates via a five-stage pipeline that instruments each step for reproducibility and deploys as an agent-callable skill. A companion tool, SetGo, further automates FAIR compliance and catalog publication, enabling raw data to become AI-ready across domains like climate and fusion.

• REDI unifies ingestion, preprocessing, transformation, structuring, and output into one reproducible pipeline.
• Agent-callable deployment allows AI agents to trigger data readiness workflows directly.
• SetGo companion tool handles FAIR compliance automation and metadata cataloging.
• Validated across climate, proteomics, materials science, and nuclear fusion datasets.

This paper examines the limitations of using local pairwise comparisons to learn human preferences for automated decision rules. It introduces a formal model for internal pluralism, where individuals hold multiple competing priorities regarding how a system should behave. The authors demonstrate that forced local comparisons fail when users cannot decisively choose between options due to these conflicting internal values.

• Pairwise comparisons assume users have a single, consistent preference for decision rules.
• Internal pluralism means people weigh multiple, potentially conflicting priorities simultaneously.
• Forced choices in alignment tasks may produce misleading data when pluralism is present.
• New evaluation methods are needed to capture complex, multi-priority human feedback.
• Designers should account for indecision as a signal of conflicting values, not noise.

Meetily is an open-source, self-hosted meeting assistant that prioritizes data sovereignty by processing everything locally. It leverages Rust for performance, offering four times faster live transcription via Parakeet or Whisper models. The tool also includes speaker diarization and Ollama-based summarization without requiring cloud services.

• 100% local processing ensures no data leaves your infrastructure, ideal for strict compliance.
• Rust implementation drives 4x faster live transcription compared to standard Whisper setups.
• Integrates Ollama for on-device summarization and speaker diarization capabilities.
• Supports macOS and Windows with a self-hosted architecture for enterprise readiness.