PPMS - Submit Project Preferences

AI-Driven Formulation Development for Pharmaceutical Applications

Formulation development traditionally relies on trial-and-error approaches combined with the expertise and experience of formulation scientists. These methods, while effective, are time-consuming and resource-intensive. With the increasing availability of evidence-based …

Formulation development traditionally relies on trial-and-error approaches combined with the expertise and experience of formulation scientists. These methods, while effective, are time-consuming and resource-intensive. With the increasing availability of evidence-based knowledge in this field and the rapid advancements in artificial intelligence (AI), we hypothesise that it is possible to predict optimal combinations of formulation excipients and methodologies for drugs using an AI-based model. Natural language processing models such as ChatGPT have demonstrated promising capabilities in prediction and success in specific applications. However, the development of such models requires extremely large datasets for training. While large language models (LLMs) are already available, task-specific supervised learning has captured significant attention. Nevertheless, fine-tuning LLMs for specific tasks presents challenges, as these models are continuously evolving. Furthermore, fine-tuning requires domain expertise to label data sourced from relatively scarce and fragmented repositories. This project aims to address these challenges by fine-tuning a suitable LLM for formulation development, focusing on: • Optimisation of excipient selection • Recommendation of candidate formulations • Prediction of formulation outcomes After developing the model, the project will focus on applying the optimised AI model to formulate two challenging drugs. These formulations will be refined using a Design of Experiments approach and subjected to in vitro assessments to ensure compliance with pharmaceutical standards. Furthermore, taste evaluation studies will be conducted with healthy volunteers to evaluate the acceptability of the medicines.

Client: Okhee Yoo

Location: Curnow building

Multiple Teams

IP: Creative Commons (open source)

Explainable Natural Language Query Interface for Relational Databases Using a Multi-Agent System

Traditional database query systems require deep technical knowledge, which is a barrier for non-expert users. Recent works tend to use machine learning methods to address this - given a natural …

Traditional database query systems require deep technical knowledge, which is a barrier for non-expert users. Recent works tend to use machine learning methods to address this - given a natural language query, the model returns the answer directly. However, these approaches are unexplainable, making them unsuitable for enterprise use where trust and transparency are critical. Given the advancement of LLMs and multi-agent systems, we can now build intelligent database exploration tools that are both powerful and explainable across diverse database schemas. Multi-Agent Architecture: The system employs specialized agents working collaboratively. For example, a Schema Intelligence Agent that automatically maps database tables and relationships into structured formats (such as knowledge graphs), a Query Understanding Agent that processes natural language queries, a Table Relevance Agent that identifies relevant tables, and a Record Discovery Agent that finds relevant records within those tables. Deliverable: You'll build a web dashboard or other visualization tools where users input natural language queries and receive relevant tables with highlighted records. If possible, accompanied by clear explanations of why each component was selected.

Client: Sirui Li, Siwen Luo

Location: Murdoch University and The University of Western Australia

Multiple Teams

IP: Creative Commons (open source)

Using AI to Grade Student Exams: Does It Work?

As generative AI technology develops, its application is expanding across various fields, with AI-assisted grading becoming an increasingly relevant topic in education. This raises important questions about the effectiveness and …

As generative AI technology develops, its application is expanding across various fields, with AI-assisted grading becoming an increasingly relevant topic in education. This raises important questions about the effectiveness and reliability of automated assessment systems: Can AI grade student work fairly and consistently? Can AI provide meaningful feedback that helps students learn? What are the strengths and limitations of AI grading systems? What role should human instructors play in AI-assisted assessment? What are the concerns about ethics and security? This project examines these questions using data from a Data Warehousing final examination administered in 2022. We will analyse and compare AI-generated grades with traditional human grading methods to evaluate the effectiveness of automated assessment tools. Research Objectives: Assess the fairness and consistency of AI grading compared to human evaluation Evaluate the quality and usefulness of AI-generated student feedback Identify key strengths and limitations of AI grading systems Determine optimal approaches for integrating human oversight with AI assessment tools Through this comparative analysis of real examination data, this study aims to provide insights into best practices for implementing AI grading systems in educational settings while maintaining assessment quality and student support.

Client: Sirui Li

Location: Office: CSSE: 224.

IP: An agreement to joint exploitation of any IP that is created

Modernising infectious disease data governance using Azure and Azure Databricks

The Department of Health, Communicable Disease Control Directorate (CDCD) manages the WA Notifiable Infectious Disease Database (WANIDD) which receives daily data from laboratories, clinics and doctors regarding patient data on …

The Department of Health, Communicable Disease Control Directorate (CDCD) manages the WA Notifiable Infectious Disease Database (WANIDD) which receives daily data from laboratories, clinics and doctors regarding patient data on infectious diseases. WANIDD is 25 years old and lacks modern cybersecurity and data governance controls. This proposal is to build a pilot system to verify whether data controls and governance could be modernised using Microsoft Azure, Azure Databricks and Unity Catalogue. Students would be provided access to the CDCD development Azure environment to set a data ingestion process for a replication WANIDD database on Azure, then using Azure tools such as Databricks and Unity Catalogue, demonstrate if and how data governance could be improved. Students would need to consider: - the problem statement including databases, stakeholders and technical environment - the proposed solution and how it would solve the problem - the high-level technical design showing which services are proposed and how they would interact - the technical build of the proposed solution, in a development environment - what would be required if the pilot solution were to be adopted for live use

Client: Chris Dorrian

Location: 189 Royal Street, East Perth

NDA Required

IP: IP to be assigned to the project proposer(s)

Ocean Genomes data visualisation

This project aims to generate creative visualisations to summarise the progress of Ocean Genomes alongside global efforts to generate high-quality reference genomes for marine vertebrate species. Students will collect, clean, …

Assessing eDNA-based sightings automatically

In this project, the team will utilise advanced algorithms to analyse environmental DNA samples collected from various habitats, from coastal environments to the deep sea. One challenge is curation of …

Geodata extraction from PDFs

To increase the value provided to its member, the Centre for Exploration Targeting (CET) at UWA, is developing a geodatabase with another partner. Parts of the data is contained in …

Bridging Structured and Semantic Data: A Framework for Transforming Relational Databases into Knowledge Graphs

In today's data-driven world, organizations face the challenge of unlocking deeper insights from their existing data assets. While relational databases have served as the backbone of enterprise data management for …

In today's data-driven world, organizations face the challenge of unlocking deeper insights from their existing data assets. While relational databases have served as the backbone of enterprise data management for decades, their rigid tabular structure often limits the ability to capture and query complex relationships between entities. Knowledge graphs, with their flexible graph-based representation, offer a powerful alternative that can model intricate connections, support semantic reasoning, and enable more intuitive data exploration. This fundamental shift from structured rows and columns to interconnected nodes and edges represents not just a technical transformation, but a paradigm change in how we conceptualize and interact with data. Deliverable: You'll develop a comprehensive framework that automates the transformation of relational databases into semantically rich knowledge graphs. The framework could employ specialized agents working collaboratively to handle the complex challenges of schema mapping, relationship preservation, and semantic enrichment. The resulting knowledge graph needs to be properly visualized through an interactive interface that allows users to explore the transformed data structure, with reference implementations available through platforms like yFiles (https://www.yfiles.com/) that demonstrate effective graph visualization techniques for complex data relationships.

Client: Sirui Li, Siwen Luo, Yihao Ding

Location: Murdoch University

Multiple Teams

IP: Creative Commons (open source)

AI generated image and document metadata

UWA Archives and Special Collections have a large number of images, documents and content that is unprocessed. Currently, adding descriptive metadata is time-consuming, involving staff viewing/reading through materials to identify …

Automated data extraction from coroner's reports

The Western Australian Centre for Road Safety Research (WACRSR) is a research centre at the university interested in all aspects related to road safety. Together with Main Roads, we are …

The Western Australian Centre for Road Safety Research (WACRSR) is a research centre at the university interested in all aspects related to road safety. Together with Main Roads, we are trying to determine the cause of road fatalities in WA and whether insights can be gained from analysing the coronial database. This project will use various tools (possibly, Large Language Models/Multimodal models, vision/document processing, etc) to extract structured data from coroners' reports. These reports are usually pdf-based documents, sometimes in plain text, sometimes as images, which makes it difficult to perform large-scale data analysis on them. The goal is to build a document processing pipeline that can automatically read these reports and pull out specific data into a consistent, structured format. While the system will initially be developed using general coroners' reports that are openly available, it is ultimately intended to help WACRSR analyse factors in road-related deaths. A primary requirement of this project is data security. The information in coroners' reports is highly sensitive, so the entire processing pipeline must be secure and self-contained. To achieve this, all processing must be done on local machines, and no data will be sent to external cloud services or third-party APIs. The system will rely on open-source, self-hosted LLMs and other local tools to ensure that all sensitive information remains within a secure environment at all times. A security audit of all tools is a must to ensure no packages unknowingly upload any data to 3rd parties.

Client: Matthew Albrecht

Location: Crawley Campus - 2nd floor Maths Link building (between Maths and EZone)

Multiple Teams

IP: Creative Commons (open source)

AI Powered Cyber Risk Assessment

Develop a SaaS platform / App to help teams perform risk assessment using the OWASP's STRIDE/DREAD methodology with the aid of AI/ML augmented with Threat Intelligence. As a cyber consulting …

Development of a Web-Based Knowledge Repository for Analytical Geoscience Workflows

The ARC Centre for Critical Resources for the Future (CCRF) at UWA supports researchers and PhD students working on critical mineral research using a wide range of microscopy and spectroscopy …

The ARC Centre for Critical Resources for the Future (CCRF) at UWA supports researchers and PhD students working on critical mineral research using a wide range of microscopy and spectroscopy techniques. As part of this work, researchers follow complex analytical workflows that are often shared informally or not documented in a consistent, accessible format. This makes it difficult for new users to independently learn or reproduce analyses, and places heavy reliance on expert support. A structured, centralised repository of protocols and training resources would significantly improve consistency, accessibility, as well as training efficiency across the centre. The aim of this project is to develop a web-based knowledge repository (e.g., using GitBook ) to host and organise analytical workflows, training materials, and best-practice protocols. These workflows will be based on structured approaches such as those described by Butcher et al. (2023). The platform should support features like embedded media, visual workflow tools, as well as the ability to upload, edit, and export content. Ideally, it will include modules for different techniques (e.g., SEM, EPMA, LA-ICP-MS, TIMA, SIMS), and allow researchers to navigate content easily, supporting self-paced learning and reproducible research across the CCRF network. The training material will build on our existing content, which will be expanded and refined to meet the new requirements.

Client: Alexandra Suvorova

Location: UWA, CMCA

IP: IP to be assigned to the project proposer(s)

Publication Auditing using LLMs

Microscopy Australia (MA) is a national network of university-based microscopy facilities that supports over 3,500 researchers each year (https://micro.org.au/). The Centre for Microscopy, Characterisation and Analysis (CMCA) at UWA is …

Microscopy Australia (MA) is a national network of university-based microscopy facilities that supports over 3,500 researchers each year (https://micro.org.au/). The Centre for Microscopy, Characterisation and Analysis (CMCA) at UWA is one of MA’s key nodes. As part of MA’s annual Research Highlights, CMCA conducts a publication audit to identify scientific papers that acknowledge CMCA, involve our staff, or reference the use of specific instruments. This process currently requires extensive manual review of journal articles and/or other scientific publications using UWA Library resources and journal databases, making it time-consuming and inefficient. This project aims to develop a semi-automated solution using Large Language Models (LLMs) to streamline the publication audit. The system will scan known scientific literature, score articles based on relevance (e.g. acknowledgements, co-authorship, instrument use), and summarise findings. A customisable user interface will allow users to configure searches, review results, and export reports (e.g. in CSV format). A self‑hosted web app solution would ensure easy access for multiple users, while a basic API would allow for integration with other services and extensibility. The outcome will improve efficiency, reduce manual effort, and enhance the visibility of high-impact research of MA node at UWA.

Client: Dr Nick Kostylev

Location: CMCA - Physics building, UWA

IP: Open to further discussion about best IP options.

Software development of visualisation and decision-making tools for future pit lake usage in a mine transitioning context

The proposed project is part of the Cooperative Research Centre program ‘Transformations in Mining Economies’ (CRC TiME) under the umbrella Project 4.9, titled "Mine Pit Lake Assessment and Management: A …

The proposed project is part of the Cooperative Research Centre program ‘Transformations in Mining Economies’ (CRC TiME) under the umbrella Project 4.9, titled "Mine Pit Lake Assessment and Management: A National Initiative to Support Mine Closure and Regional Opportunities". Project 4.9 brings together an interdisciplinary group of researchers, collectively committed to exploring beneficial use options and suitable modelling approaches for priority mine pit environments in Australia. With reference to the internship, the assigned student(s) will work with the pit lakes researchers to aid the development of a mobile app or web-based platform envisioned as a suite of tools comprising functional prototypes of software and interactive data visualisation outputs that are fit for purpose and use. Specifically, the functional prototypes are intended to facilitate better decision making about pit lake futures, by supporting informed dialogue between stakeholders about their values and the possibilities for future pit lake uses. While the research team is in the process of refining our methodology and developing the suite of tools, the software design component will need to be developed from scratch with collaborative guidance from us. Ideally, the student(s) assigned to this project will be comfortable working within a professional academic context and familiar with principles of user experience design and software engineering, with iterative coding and testing being a core component of the development process. Additionally, under the guidance of the research team, the student(s) will clearly document their methods and processes, contextualising their activities within emerging research in transdisciplinary practice.

Client: Chantal Bourgault Du Coudray

Location: UWA

IP: Open to further discussion about best IP options.

Data-Driven Discovery: 3D Gap Analysis of Imputed vs Original Copper Assays in Western Australian Drillhole Data

Mineral exploration in Western Australia is supported by vast volumes of geoscientific data, including government-hosted public datasets of exploration drillhole records. This project applies data science methods to a curated …

Mineral exploration in Western Australia is supported by vast volumes of geoscientific data, including government-hosted public datasets of exploration drillhole records. This project applies data science methods to a curated subset of this dataset, focusing on Copper (Cu) assay values. The primary objective is to compare the original laboratory assay results with a derived dataset containing imputed values for samples where Cu was not originally analysed (Deep-learning identification of anomalous data in geochemical datasets). This imputation introduces new information that could guide exploration decisions—particularly in identifying prospective zones where Cu may have been missed during the initial sampling phase. The key challenge lies in the spatial nature of drillhole data, which is inherently three-dimensional. Differences between the original and imputed datasets are not easily visualised or interpreted using traditional 2D methods (https://www.sciencedirect.com/science/article/pii/S0169136822005509?via%3Dihub). This project will trial data analysis and visualisation techniques to detect patterns and highlight zones of divergence between datasets. Exploratory data analysis, geostatistical comparisons, and 3D visualisation tools will be employed to reveal “hidden” exploration targets that would otherwise be overlooked by the human eye. While this proof-of-concept will focus on Cu, the broader aim is to design a flexible analytical platform that could extend to other commodities such as Au, Ni, or Li. Users would be able to input their own paired datasets (original and imputed) for different elements, allowing for scalable and element-agnostic geoscientific analysis. The outcome will support smarter targeting decisions in mineral exploration, combining machine-aided inference with interactive visual interpretation. Project supervisors: Paul Duuring (Geological Survey of Western Australia)

Client: Paul Duuring

Location: Mineral House, 100 Plain Street, East Perth

Multiple Teams

IP: Creative Commons (open source)

Isolating Financial Market Sentiment in Transmission of Monetary Policy

There are two broad areas in which I would like to explore applications for written software. One area is monetary policy, where I aim to utilize Large Language Models to …

Harnessing Big Data to Improve Financial Integrity and Operational Efficiency

Big data plays a transformative role in improving organisational efficiency by converting large, complex datasets into actionable insights. In higher education, where student fees, government funding, and loan programs operate …

Big data plays a transformative role in improving organisational efficiency by converting large, complex datasets into actionable insights. In higher education, where student fees, government funding, and loan programs operate across fragmented systems, big data offers a powerful solution for streamlining operations and enhancing financial oversight. By integrating disparate data sources—including university records, payments, and funding estimates—this project leverages data and analytics to improve transparency, automate validation processes, detect inconsistencies early, and reduce manual reconciliation efforts. This project explores the application of big data to enhance financial integrity and operational performance within a university setting. By linking data from internal student systems with external government platforms such as Tertiary Collection of Student Information (TCSI) and Higher Education Information Management System (HEIMS), the project enables a consolidated, bottom level details of revenue. This centralised model supports real-time reporting, automated discrepancy detection, and improved forecasting capabilities. Beyond technical outcomes, the project also contributes to institutional resilience by embedding knowledge into sustainable systems, reducing single-point dependencies, and supporting data-driven decision-making in complex funding and government policy environments. The project will begin by integrating data sources and developing data pipelines using tools such as Python and SQL to extract, cleanse and integrate this disparate data into a centralised repository designed for the analysis. Reconciliation logic, validation rules and reporting dashboards will be implemented (in Power BI) to detect discrepancies, provide interactive reporting and exception monitoring, while automation tools will schedule data refreshes and trigger alerts for anomalies. Comprehensive documentation and knowledge capture will ensure sustainable handover and continuous improvement. This project will provide you with an opportunity for practical application of your knowledge in a real-world situation, enhance your employability, and help the University improve back-office operations.

Client: Zhibo Guo

Location: UWA Crawley Campus

NDA Required

Multiple Teams

IP: IP to be assigned to the project proposer(s)

RAG Chatbot with Spatial-Temporal Data from Australia Youth Atlas

Creation of a prototype chatbot for the Australian Child and Youth Wellbeing Atlas. Led by A/Prof Bec Glauert and her team at the School of Population and Global Health, the …

Creation of a prototype chatbot for the Australian Child and Youth Wellbeing Atlas. Led by A/Prof Bec Glauert and her team at the School of Population and Global Health, the Australian Child and Youth Wellbeing Atlas is an innovative project that consolidates data on the health and wellbeing of children and young people in Australia. This includes data from WA's new linked data asset PeopleWA and many other data sets from each state and territory. In total, the Atlas contains more than 500 data sets. The Atlas is a free mapping resource designed to facilitate access to and use of this data, supporting informed decision-making for the future of our children and young people. The proposed Atlas chatbot will feature several key functions: a non-intrusive pop-up on the ACYWA main page, that accepts user enquiries for two purposes, one is to produce an answer (either numerical or summary) by retrieving the data directly from the backend datasets; the other is to direct user to the right visualisation that demonstrate the results. The chatbot should perform faithfully without hallucination. Here are the links to the Atlas and maps:

https://australianchildatlas.com/
https://australianchildatlasmap.sph.uwa.edu.au/

Client: Marketa Reeves

Location: School of Population Health, UWA

NDA Required

Multiple Teams

IP: Creative Commons (open source)

Beyond Typing: Engineering Real-time Conversational AI

The project involves the development of a framework for low latency, audio based chat interface to large language models (LLM). The aim of the project is to provide a more …

The project involves the development of a framework for low latency, audio based chat interface to large language models (LLM). The aim of the project is to provide a more natural way of conversing with LLMs. The user should be able to initiate a conversation with the press of a button. The app captures the users’ voice using a microphone on the device, converts it to text and sends it to a language model or custom text-to-speech (TTS) model to provide a (near) instant audio response. ChatGPT or Gemini can create a web app that can do this in less than 10mins, but the experience is poor with high latency and unnatural TTS. The key challenge in this project is to leverage and integrate state of the art open source frameworks in a way that minimises latency and provides the most natural conversational interface possible. Ideally, the app can recognise pauses in speech and pick the right time to respond without the user having to explicitly press a send button. The engine should use advanced techniques like input streaming, key-value cache management and/or prefilling/speculative sampling to reduce the latency of the response. Creative use of gap fillers can also be used in the response to compensate for delays in the pipeline. It is expected that the development team can quantify performance on different hardware platforms using benchmarking techniques to progressively refine the experience. The state of the art in this space is sesame.com. Their technology is crossing the uncanny valley. They have sample code for their CSM (Conversational Speech Model) on github (https://github.com/SesameAILabs/csm/tree/main) which can provide a starting point for this exploration. The project should also include research into other key (open source) enabling technologies in this rapidly evolving field. In the near future we will be discussing the news or the latest Netflix series with LLMs in the same way as we converse with our friends and family. This project aims to push the boundaries of what's possible with open-source frameworks, tackling challenges in low-latency streaming and async pipelines, and innovating with methods like gap fillers to achieve highly realistic conversational experiences.

Client: Julien Flack

Location: 121 St Georges Terrace, Perth, WA 6000

Multiple Teams

IP: Creative Commons (open source)

View Projects or Submit Project Preferences

CITS5553 - Data Science Capstone Project

Group Information

Only use this form for a team's preferences

Your Selected Projects

Only use this form for a team's preferences

How to Rank Projects:

Available Projects

AI-Driven Formulation Development for Pharmaceutical Applications

Explainable Natural Language Query Interface for Relational Databases Using a Multi-Agent System

Using AI to Grade Student Exams: Does It Work?

Modernising infectious disease data governance using Azure and Azure Databricks

Ocean Genomes data visualisation

Assessing eDNA-based sightings automatically

Geodata extraction from PDFs

Bridging Structured and Semantic Data: A Framework for Transforming Relational Databases into Knowledge Graphs

AI generated image and document metadata

Automated data extraction from coroner's reports

AI Powered Cyber Risk Assessment

Development of a Web-Based Knowledge Repository for Analytical Geoscience Workflows

Publication Auditing using LLMs

Software development of visualisation and decision-making tools for future pit lake usage in a mine transitioning context

Data-Driven Discovery: 3D Gap Analysis of Imputed vs Original Copper Assays in Western Australian Drillhole Data

Isolating Financial Market Sentiment in Transmission of Monetary Policy

Harnessing Big Data to Improve Financial Integrity and Operational Efficiency

RAG Chatbot with Spatial-Temporal Data from Australia Youth Atlas

Beyond Typing: Engineering Real-time Conversational AI

Ranking Guidelines

Consider Team Skills

Evaluate Requirements

Rank Strategically