Computer science student at the University of Melbourne.
Building AI agents and practical software, with a focus on reliable answers, explicit workflows, and useful interfaces.
I am interested in the gap between an impressive model response and a dependable product. My projects explore how retrieval, state, tools, and interface design work together to make AI systems useful.
The problems I keep working on:
- Evidence-grounded answers: retrieving relevant information, checking sources, and recognising when there is not enough evidence to answer.
- Controlled agent workflows: combining flexible model behaviour with explicit steps, validated inputs, and clear failure paths.
- State and recovery: keeping user data consistent across long conversations, interrupted requests, and retries.
- Human-facing software: making complex systems understandable through clear interfaces, explanations, and accessible interactions.
| Project | What I am building |
|---|---|
| LeapAshore | A master's application-planning agent covering the QS 2027 top 300: a six-phase roadmap, rule-by-rule reach / match / safe tiers, and a LangGraph advisor that passes an evidence gate before answering and asks before changing the school list. Deployed; most admission requirements are still being collected and are marked as pending verification. |
| StudyOrbit | A local-first UniMelb coursework assistant: reads Canvas, Ed and the Handbook from the signed-in session, has a model turn them into validated course JSON, then shows grades, plans the week and answers questions about each course. Still a prototype. |
| SageSense | An Android anti-scam companion built with a team: local risk detection, bilingual explanations, and a citation-backed advisor. |
AI / backend
Web / applications
Data / workflow
Contribution visualisation: github-profile-3d-contrib.



