Research · Université de Montréal, DIRO

AI for Software Engineering

I study the reliability, cost and trade-offs of AI coding agents, comparing open-source and proprietary toolchains, with a focus on automated program repair and how we evaluate it.

View CV Collaborate
  • Université de Montréal
  • GEODES
Members of the GEODES software engineering lab on a dock by a lake
GEODES lab, Université de Montréal, 2026

Interests

What I work on

  • 01

    AI coding agents

    Empirical analysis and comparative evaluation of open-source and proprietary coding agents: reliability, cost, and trade-offs.

  • 02

    Automated program repair

    Context retrieval and feedback-driven patch generation for agentic repair pipelines.

  • 03

    Evaluation & benchmarks

    Benchmark design and LLM-as-judge methods for developer tooling, beyond "the tests pass".

Publications

Publications & manuscripts

Experience

Research experience

  1. Feb 2026 – Aug 2026Montréal, QC, Canada

    Software Engineering Research Intern

    Université de Montréal

    • Reviewed automated program-repair pipelines and identified research gaps in context usage and feedback-driven patch generation.
    • Proposed context retrieval for agentic program repair that preserves repair performance while using 14% fewer tokens.
    • Applied cross-task knowledge distillation to improve LLM patch generation from patch-validator feedback.
    • Developed an LLM-as-Jury to detect patches that pass tests but are incorrect or fail in production.

Toolkit

Methods & tools

ML & AI
PyTorch, HuggingFace, Scikit-learn, TensorFlow, LLM Evaluation, Knowledge Distillation, Agentic AI Systems
Research methods
Benchmark Design, Experimental Design, Empirical Evaluation, Data Visualization
AI coding agents & SE
Automated Program Repair, SWE-bench, OpenCode
Communication
LaTeX, BibTeX, Scientific Writing, Technical Presentations