Crii: OAC: aquillm: a knowledge-based AI tool to keep research data alive

This federal Project Grant award of $164,123, provided by the National Science Foundation's (NSF) Computer and Information Science and Engineering (CISE) program, funds the development of AQUILLM - a knowledge-based AI tool to help research groups better manage and access their data.

The AQUILLM system uses retrieval-augmented generation and large language models to enable natural language querying of private research data such as papers, experimental results, and meeting notes. This allows research groups to efficiently discover and access information across their growing data repositories, especially as researchers join and transition out of projects over time. The project leverages open-source tools to create a scalable, secure, and customizable solution that protects data privacy while providing an intuitive interface for research data management. The award supports the development of this system at Southern Oregon University over the 2-year period from July 2025 to June 2027.

Abstract

The introduction of retrieval augmented generation and large language models (RAG/LLM) shows promise as a way for scientific research groups to discover and access information across platforms using natural language queries. Research groups often face challenges managing and accessing work such as paper drafts, research experiments, plots, and meeting notes, especially as these resources grow over time and researchers transition in and out of projects. This project addresses these challenges by developing aquillm, a RAG/LLM tool that leverages open-source models and platforms, and serves as a knowledge base for research groups by protecting private data, preserving tacit group-specific knowledge, and offering an open-source design to allow other researchers to implement and customize their systems with careful attention to cost considerations. This project benefits research groups by offering a way to use natural language to ask questions about their data, yielding links to relevant documents. Aquillm is designed to address the challenge of securely organizing and accessing research group data using natural language queries. The system consists of a django-based web interface connected to a vector store for indexing and retrieving data efficiently. IT integrates with a large language model (LLM) to process queries and return relevant results in real-time. The project uses open-source tools to ensure adaptability and scalability while maintaining data privacy. Researchers can query private datasets, such as research papers, experimental results, plots, and meeting notes, without exposing this data to external systems. Aquillm employs modular design principles, allowing components like LLM or vector store to be swapped or customized for specific research group needs. Its contribution lies in providing a scalable, secure, and customizable solution for integrating RAG/LLM systems into research environments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.