CV

Curriculum Vitae

📄 Download 1-Page PDF CV

Professional Profile

AI Researcher specializing in machine reasoning and alignment. Blends formal logic with rigorous low-level systems engineering to architect custom agentic frameworks and highly creative adversarial evaluations. Former Military Intelligence Operator with a Top Secret clearance, bringing a rigorous, threat-modeling mindset to AI security and model capability evaluations.

Technical Skills

• Languages: Python, C++, Java, SQL, LaTeX
• ML & AI: PyTorch, vLLM, HuggingFace, Transformers, Google/Anthropic/OpenAI APIs, Agent Harness
• Systems & Infrastructure: Linux/Ubuntu Server, AWS, Apache Flink, Distributed GPU Clusters
• Core Competencies: Large Language Models (LLMs), Machine Reasoning, AI Alignment, Context Management, Mechanistic Interpretability, Adversarial Evaluations, Threat Modeling, Distributed Systems

First-Author AI Research

• ArbiGraph (2026 – Present) [arXiv] | [GitHub]
  • Built ArbiGraph, a Python evaluation framework for testing whether LLM agents can follow long, multi-step workflows without losing, mixing up, or reusing stale intermediate state.
  • Converted math, word-problem, and Python-tracing tasks into automatically generated workflows with executable ground truth, enabling exact grading without manual labels.
  • Added controls for workflow length, branching, irrelevant distractors, and value types, letting researchers reproduce agent failure modes and scale difficulty without hand-crafting prompts.
  • Implemented the agent evaluation harness around a calculator tool, including answer extraction, tool-call validation, continuation handling, and repair prompts for incomplete or malformed runs.
• Robust Reasoning Benchmark (RRB) (2026) [arXiv] | [DOI] | [GitHub]
  • Designed a highly creative adversarial evaluation framework leveraging 13 deterministic textual perturbations to decouple an LLM’s mechanical deciphering from its underlying mathematical logic.
  • Demonstrated that the attention drift occurs within a single query’s Chain-of-Thought, empirically showing that intermediate reasoning steps pollute the dense attention mechanism, identifying the optimal granularity of reasoning as a critical open research problem.
  • Engineered custom mechanistic interpretability pipelines in PyTorch to extract and analyze causal attention probability matrices across token index boundaries, testing models ranging from 7B to 30B parameters.
• Fusing Adds and Shifts for Efficient Dot Products (2026) [IEEE CAL] | [GitHub]
  • Proposed and validated a novel algorithmic optimization for dot-product computations, demonstrating a strong foundational understanding of hardware-level ML primitives and efficiency.

Engineering & Professional Experience

• Systems & Infrastructure Engineering (MSc Thesis) (2020 – 2022) University of Toronto
  • Engineered a flexible IoT distributed data-streaming framework from scratch, designed to automatically partition computational streaming queries between edge devices and cloud instances.
  • Built the full software stack: programmed Arduino/C++ sensors for real-time biological data collection (EMG/ECG), developed custom socket networking protocols, and deployed cloud infrastructure using AWS and Apache Flink.
• Intelligence Operator (2013 – 2018) Canadian Armed Forces
  • Held a Top Secret security clearance, conducting rigorous analysis of classified information streams to produce actionable intelligence reports for command elements.
  • Developed a strong adversarial threat-modeling mindset, emphasizing operational security, rigorous data validation, and the identification of logical vulnerabilities in complex, multi-agent scenarios.
• Mathematics Teacher (2012 – 2013; 2018 – 2019) Blyth Academy
  • Taught foundational mathematics to students in Grades 10, 11, and 12, developing the ability to distill and communicate complex quantitative concepts.

Co-Authored Systems Research

• GPUPool: A Holistic Approach to Fine-Grained GPU Sharing in the Cloud PACT 2022 | Co-authored with Xiaodan Serina Tan, Nandita Vijaykumar, Gennady Pekhimenko.
• Habitat: A Runtime-Based Computational Performance Predictor for Deep Neural Network Training USENIX ATC 2021 | Co-authored with Geoffrey X. Yu, Yubo Gao, Gennady Pekhimenko.

Education

• PhD in Computer Science (Paused to transition to industry), University of Toronto, 2022 – Present
• Master of Science (MSc) in Computer Science, University of Toronto, 2020 – 2022
• Bachelor of Science (BSc) in Mathematics and Philosophy (Formal Logic), University of Toronto, Graduated 2011