Gravitational Lens Detection
ResNet-18 models for strong-lens detection under 100:1 class imbalance, for ML4SCI / DeepLense.
0.9877 AUC · 96.9% sensitivity · 0.9919 AUC on substructure classification
PythonPyTorch
About
I'm an M.Sc. student in Computer Science at UBC, supervised by Prof. Cristina Conati in the HCI-NLP Lab and working with Dr. John-Jose Nunez on human-centred AI for cancer care with BC Cancer.
My research looks at what language models are doing internally and how their outputs should be explained to the people who rely on them: probing hidden states, patching activations, and designing explanations that hold up in sensitive settings like oncology.
I also enjoy low-level systems work. Most recently I built the global task registry for the Swift Concurrency runtime through Google Summer of Code, and before that I wrote flight software for a rocket team and shipped production web apps for a professional football club. I did my B.Sc. in Computer Engineering at Politecnico di Torino.
Research interests
- LLM analysis & interpretability
- Hidden-state probing & activation patching
- Evaluation of language models
- Retrieval-augmented generation
- Explainable AI
- Human–AI interaction
News
- Nov 2026Upcoming: stakeholder workshop on explainable cancer-care chatbots at the BC Cancer Summit (19 Nov).
- Sep 2026Started my M.Sc. at UBC, supervised by Prof. Cristina Conati and working with Dr. John-Jose Nunez.
- Sep 2026Graduated from Politecnico di Torino with a B.Sc. in Computer Engineering (101/110).
- Sep 2026Wrapped up Google Summer of Code on the Swift Concurrency runtime.
Research
- Manuscript in preparation2026 – present
Explainable AI for Cancer Care Navigation
UBC × BC Cancer
What explanations should a cancer-navigation assistant give, and in what form? Informed by a 16-paper review of healthcare XAI, trust, empathy and personalised explanations, plus participant data from the BC Cancer Summit 2026. Compares why / how / no-explanation options, text, visual and conversational formats, and direct versus supportive-resource recommendations.
- In progress2026 – present
LLM Forgetting: Hidden-State Probing and Activation Patching
with Finn Vamosi
How do language models behave when instructed to forget information? We use hidden-state probing and activation patching to test how interventions at selected activations relate to later behaviour.
- In progress2026 – present
Subliminal Learning from Weight-Finetuned Teachers
with Finn Vamosi, Erell Boutin and Zung Nguyen
Can behavioural traits of a fine-tuned teacher model emerge in a student? We study how weight updates and LoRA configurations relate to that transfer.
- Public report2026
Analysis of AI Usage in University-Level Lecturing
with Alp Kale and Bora Demirtola
Evaluated AI systems for lecturing and instructional-material generation, including a full Computer Architecture lecture.
- Course report2026
AI-Based Anomaly Detection in Network Security
with Alp Kale and Demir Ege Bulte
Investigated machine-learning approaches to detecting anomalous activity in network-security data.
Experience
Graduate Research Assistant
Sep 2026 – PresentUBC & BC Cancer · Vancouver, BCResearch
- Contribute to an LLM- and RAG-based cancer-care navigation assistant connecting users with oncology and supportive-care information.
- Study how recommendations should communicate evidence: why a recommendation was made, how it is grounded, and when to suggest further resources in sensitive situations.
- Helped design a two-hour stakeholder workshop for the BC Cancer Summit combining hands-on chatbot sessions with round-table discussion among patients, caregivers, clinicians and researchers.
Google Summer of Code Contributor
May 2026 – Sep 2026Swift · Concurrency Runtime · RemoteEngineering
- Implemented a 64-shard global task registry in the Swift Concurrency runtime so suspended async tasks are discoverable by debuggers and crash-analysis tools.
- Designed concurrent C++ data structures and synchronization paths, keeping overhead below 5% on the Swift benchmark suite.
- Extended LLDB Python tooling for live task enumeration and contributed to signal-safe crash-backtrace integration.
Student ML Engineer & Researcher
Feb 2026 – Jul 2026Politecnico di Torino · Turin, ItalyResearch
- Developed models predicting student academic trajectories from institutional data, from feature engineering and cohort analysis to evaluation.
- Built data-processing and evaluation pipelines for large tabular datasets and supported integration of AI class-support tools.
Software Engineer Intern
Sep 2025 – Feb 2026Parma Calcio 1913 · Parma, ItalyEngineering
- Built production React/TypeScript applications for medical-staff workflow management and operational tooling.
- Implemented AWS S3/CDN asset delivery and real-time data pipelines, cutting load times by 80% with Lighthouse scores above 99.
Embedded Software Engineer
Feb 2025 – Feb 2026PoliTO Rocket Team · Turin, ItalyEngineering
- Wrote STM32 flight software in C/C++: peripheral drivers, watchdogs, fault handling and SWD debugging under real-time constraints.
- Built internal tools used by 150+ team members, reducing administrative overhead by 95%.
Projects
DAG-based simulation engine for merge, rebase, reset and cherry-pick with a live commit-graph view. Published on PyPI.
135+ integration tests · 3rd place, GitKraken Game Jam
PythonCI/CDPyPI
Privacy-first macOS system monitor that runs a reconstruction autoencoder on the Apple Neural Engine to flag memory leaks.
SwiftCore ML
Real-time WebSocket chat server built on SwiftNIO, focused on high-performance async networking.
SwiftSwiftNIO
Education & Honours
University of British Columbia
Sep 2026 – Jun 2028M.Sc. in Computer Science (thesis-based) · Vancouver, BC
- Supervised by Prof. Cristina Conati; also working with Dr. John-Jose Nunez.
- Coursework: Analysis of NLP Models, Open Questions in NLP, Machine Learning Theory, AI Evaluation, Reinforcement Learning, Human-Computer Interaction.
Politecnico di Torino
Sep 2023 – Sep 2026B.Sc. in Computer Engineering · Turin, Italy
- GPA 3.7/4.0 (27.5/30), final mark 101/110. Full merit scholarship.
- Coursework: Computer Architecture, Algorithms and Data Structures, Operating Systems, Object-Oriented Programming, Programming Techniques, Computer Networks, Introduction to Databases, Introduction to Data Science and Visualization, Digital Electronics Design, Signal Analysis and Processing, Automatic Control, Cybersecurity and National Defence, Linear Algebra and Geometry, Mathematical Methods for Computer Science, Mathematical Analysis I and II.
Honours
- Top 9 finalist, Amazon Robotics Hackathon
- 3rd place, GitKraken Game Jam (GitSimulator)
- KU Leuven CS Scholarship, selected among the top ten incoming students
Skills
- ML & Research
- PyTorch, scikit-learn, LLM evaluation, probing & activation patching, RAG, XAI, computer vision
- Languages
- Python, C, C++, Swift, TypeScript, Java, SQL
- Systems & Tools
- Linux, LLDB, STM32, React, Next.js, PostgreSQL, AWS, Docker, CI/CD
Outside the lab
- Music: Learning guitar on a Yamaha Pacifica through a Boss Katana Mini.
- Caffeine: Iced coffee with milk, almost exclusively.
- Activity: Working out and running around Vancouver.
- Languages: Turkish, Italian, English, and slowly, French.
