Projects

Rubric-Based Intelligent Curation (RUBICON) is a modular framework for building sailor-centric, domain-specific systems that curate and guide text-based reporting at the time of entry. It uses large language models (LLMs), report quality rubrics (RQRs), and an automated initialization curriculum that can be adapted to a wide range of report domains with minimal human input. Using an interactive in-line suggestion interface, RUBICON simultaneously reduces the warfighter’s burden for detailed report writing while improving report data quality.

ChatBS: A Context-aware LLM Exploratory Sandbox uses the OpenAI Completion API service (GPT-4-0613 model) to answer questions. Each sentence in a ChatBS result is automatically linked to a Google query to facilate fact-checking. If requested, ChatBS can then use the OpenAI API to construct an entity/relation graph of these results in the form ['entity1', 'relationship', 'entity2'].

Whyis is a provenance-aware nano-scale knowledge graph publishing, management, and analysis framework. Whyis aims to support domain-aware management and curation of knowledge from many different sources. Its primary goal is to enable creation of useful domain- and data-driven knowledge graphs. Knowledge can be contributed and managed through direct user interaction, statistical analysis, or data ingestion from many different kinds of data sources.

Semantic Extract, Transform, and Load-er (SETLr) is a flexible, scalable tool for providing semantic interpretations to tabular, XML, and JSON-based data from local or web files. It has been used by diverse projects and has shown to be scalable and flexible, allowing for the simplified creation of arbitrary RDF, including ontologies and nanopublications, from many different data formats. Semantic ETL scripts use best practice standards for provenance (PROV-O) and support streaming conversion for RDF transformation using the JSON-LD based templating language, JSLDT.

Automated clusteRing Curriculum LearnIng Guided by Human Training (ARCLIGHT), is a classification engine capable of (1) automated discovery and characterization of objects and activities in multimedia data and (2) solicitation of input from human analysts to refine, correct, or update its internal knowledgebase.