Curtin University: Helping Thousands of International Research Students Find Answers – Instantly, Securely, In Any Language

Brief

We built a fully offline AI chatbot for Curtin University that gives HDR students fast, accurate, source-backed answers, in whatever language they ask. Powered by Qwen 3 running entirely on Curtin’s own hardware, the system searches hundreds of official PDFs and webpages, always cites where the answer came from, and never sends a byte of data outside the university. The whole thing runs without external APIs, cloud costs, or privacy risk.

We highly commend the AI team at AI Advancements. What stood out most was their clear and responsive communication, as well as their genuine dedication to customer satisfaction.

Sonny Pham, Associate Professor at Curtin University

Problem

HDR students at Curtin are some of the most capable researchers in the country, but even the sharpest minds hit a wall when the information they need is buried across multiple websites, policy documents, and support portals, each updated on its own schedule and written for a different audience. For students whose first language isn’t English, the problem compounds fast: a question about milestone submissions or visa compliance might take hours to answer properly, not because the information doesn’t exist, but because finding it and trusting it is genuinely difficult. Curtin had already tried a GPT prototype, but it required manual document uploads, had no verification layer, and couldn’t confirm whether its answers were actually drawn from official sources. They needed something that could search a large body of official content, always cite the source, work in any language, and keep every piece of student data inside Curtin’s own walls.

Solution

We built a complete locally hosted chatbot ecosystem from the ground up. Qwen 3 runs on Curtin-owned hardware, meaning no data ever leaves the university. A custom Retrieval-Augmented Generation pipeline indexes content from official PDFs and scraped webpages, so every answer is drawn from a verified source, and students can click the reference link to confirm it themselves. If no sufficiently relevant source exists, the chatbot doesn’t guess. It redirects students to the right support channel instead.

• Fully offline, fully on-premises: the LLM, the data, and the interface all run on Curtin infrastructure, with no API calls, no cloud costs, and no data leaving the university
• Source-linked answers: every response cites the exact document or webpage it drew from, so students and administrators can verify accuracy at any time
• Multilingual by design: Qwen 3 handles queries in any language, so international students can ask in the language they think in
• Built to grow: the document ingestion pipeline is designed so the research team can add new content as Curtin’s policies evolve, without rebuilding anything
• Accuracy-first behaviour: when the system isn’t confident enough, it says so and points students to a human, with no hallucinated guidance on compliance-critical questions

Result

HDR students now have a fast, reliable path to accurate information, in their language, available whenever they need it. Early feedback shows students are finding answers more easily and spending less time hunting across scattered documents and support pages. For the research team, the chatbot provides a secure, extensible foundation for ongoing research into how AI can improve the HDR student experience. The offline architecture means Curtin controls the data, controls the costs, and has a platform that can scale as more content is added. It’s the kind of thing that looks simple from the outside, ask a question, get an answer, but the reliability, privacy architecture, and multilingual capability underneath are what make it genuinely useful in a high-stakes academic environment.

Under the Hood