Ask My Business, a deployed RAG widget answering customer questions for small businesses. Jebb, the concierge answering questions on this page. A capstone that reproduced five few-shot CLIP methods instead of picking one.
six projects, ordered by weight, not date.
Ask My BusinessLive
An embeddable AI chat widget for small businesses.
Are you open on Saturdays?Yes, 8am to 1pm. Walk-ins are welcome, bookings get priority.Illustrative example
Jebb
My personal AI operating system.
Few-Shot Adaptation of CLIP (Capstone)
Reproduced and compared five state-of-the-art few-shot adaptation methods for CLIP: LP++, CoOp, PromptSRC, TaskRes, and PromptKD.
Legal Risk RAG (confidential)
Confidential technical contribution to an Australian legal-tech startup.
AutoLZ
A computer vision pipeline for detecting safe landing zones for autonomous drones.
SonarMapper
A PPO-based reinforcement learning agent for autonomous underwater navigation in a simulated environment under partial observability.
Jebb answers questions about this work, in Aryan's voice.
Finishes things.
Bachelor of Artificial Intelligence graduate from UTS (2026, WAM 78.88). Builds applied AI systems end to end: a deployed multi-tenant RAG chat system, a personal AI assistant platform used daily, and a technical contribution to an Australian legal-tech startup. Capstone on few-shot adaptation of CLIP was awarded a High Distinction. Currently exploring both industry roles and research directions in perception for autonomous systems.
Junior roles in AI/ML engineering, computer vision, robotics, or autonomous systems. Open to product-shipping teams and to research-adjacent engineering roles. Sydney or Australia-wide preferred; remote considered.
Human Cognitive State Monitoring via Brain Connectivity Exploration and Graph Neural Networks
detecting driver drowsiness from EEG signals before an incident occurs. Standard approaches treat each electrode independently or use simple spectral features. This project investigates whether the connectivity structure between brain regions, modelled as a graph and processed by graph neural networks, carries information that electrode-level features miss, and specifically whether directed connectivity (which region drives which) outperforms undirected coupling metrics.
Brain-computer interfaces as an input modality for autonomous systems (drones, prosthetics, eventually space robotics) sit at the intersection of perception, learning under uncertainty, and control.