As more data becomes available, developing easy-to-use tools for searching data offers a significant opportunity to improve accessibility and engagement. Artificial intelligence (AI) chatbots that use Retrieval-Augmented Generation (RAG) can help users quickly ask questions about data and engage with research results in a simple, conversational way. This technology can streamline workflow, reduce the time spent on finding data and improve decision-making by giving relevant answers. This will lead to greater engagement with the Centre's collection of research outputs.
Given the opportunities, risks and constraints discussed in the RFP, the Centre would like to evaluate the feasibility of RAG implementation within our information and technology systems. The main objectives are to:
- provide a background of RAG as a technology
- outline any non-technological issues or constraints
- identify potential use cases
- develop at least three options for implementation
- compare these options using SWOT analysis and cost/time estimates.
Frequently asked questions
Q) Why is DeepSeek of specific interest over other LLMs (especially when the Federal Government has requested no use of DeepSeek and by the time the contract is signed there could be another model available)?
A) The Centre is not a government entity, although we seek to align with the Commonwealth Government where feasible (given our best interests).
DeepSeek was only mentioned as an example due to it being one of few open-source, high-performing LLMs which can operate on consumer-grade, relatively inexpensive hardware and is permissively licensed under MIT.
The Centre is open to additional options beyond the three outlined (these are the minimum).
Q) What types of source materials should we include? For example, government policy papers, technical reports, peer‑reviewed articles, and books on natural hazards—should we also incorporate tables, figures or image data? Are these data types updated real time?
A) The Centre works with a broad variety of Research Data Assets (including, but not limited to maps, text documents, datasets, data sources, images and videos).
The specific answer to this question is part of the requirements of this work: “3. Identifies potential use cases within the Centre…”
Q) Beyond the initial feasibility study, is it expected that the project will progress to a full chatbot development and deployment phase over one to three years? Or longer?
A) The answer to this question requires the completion of this work.
Q) To what extent should the project team demonstrate prior experience with Retrieval‑Augmented Generation (RAG) techniques?
A) Prior experience with RAG will be assessed under the “Capability” criterion. You should demonstrate to the best of your ability any experience with RAG or similar technologies.
Q) Does the feasibility study need to showcase how RAG and large language models perform when applied to a limited corpus of documents?
A) We cannot answer this question as it may be related to your proposal.
Q) I did not find your application from for the feasibility studies from your website. Do you have the application form?
A) No, we do not have a template for this project submission however the response should contain no more than 10 pages (including appendices or other attachments). Excluding the mandatory sections - the format has been left intentionally flexible. This is opportunity to demonstrate your understanding and communication skills with us.
Q) Was there an EOI prior to this RFP?
A) No, this is not a core research project.
Q) Can you confirm what the existing technology environment is at NHRA (e.g. Microsoft/AWS/GCP) and is the intention for solutions to align with this stack?
A) The scope of this work is a feasibility study, not the development of software. While the technology stack may be a detail of the study, it is not necessary within the context of a proposal for the study.
