Blog ·
GitHub ·
Scholar ·
LinkedIn ·
Twitter

Zarreen Reza

I study how AI systems behave, and how they quietly reshape the people who use them.

Over six years I have moved from applied ML in industry to research on AI safety,
interpretability, and human-centric AI. My recent work, with Marwa Abdulhai and Natasha Jaques, examines how AI assistants homogenize human writing and inject their own tone, why personalized agents fail to track changing preferences, and what post-training changes inside a model, aiming to make hidden AI effects measurable through large-scale study, controlled experiments, and interpretability.

I am a Senior AI Research Engineer at JoulesAI, applying reinforcement learning to large-scale sequential decision-making and constrained optimization (PPO agents for energy dispatch under real cost, carbon, and operational constraints), and a Fatima Research Institute Predoctoral Fellow in mechanistic interpretability.

The question I most want to work on: how do we align AI with human goals, not just task performance?

Publications

(*equal / co-first, †alphabetical /
random authorship)

Dissertations & Selected
Research Reports

Other Research

Experience

Teaching

• Instructor & content creator, OpenMined
  Private AI Series on Federated Learning (funded by Meta AI, PyTorch, Oxford), 2021.
• Tutor, COMP-8547 Advanced Computing Concepts, U. Windsor, 2018–2019.
• Graduate TA, U. Windsor (60-256 Systems Programming, 60-141 Algorithms & Programming II,
  60-254 Data Structures), 2017–2019.
• TA, BRAC University (CSE110/111 Programming I–II, CSE220 Data Structures, CSE260 Digital
  Logic, CSE221 Algorithms), 2013–2016.

Talks

Service

• Leadership Fellow, Women Who Code (2021–2022): led the Data Science Track of 4,500+
  members and 50+ volunteers across 150+ events; spoke at 40+ events.
• Organizer: WWCode DS Career Mentorship 2022, ML Study Group 2022, BlockDataPy Summit 2022, Hackathon for Social Good 2022.
• Reviewer / Judge: CaLM Workshop, NeurIPS 2024; TigerGraph Graph
  for All, 2022; WRSTEF,
  2019.
• Mentor: OpenMined Beta
  Bootcamp (2020); Technovation Challenge (2018).

Mentorship

• JACOBB: led an 8-member AI team; mentored five junior researchers on code
  quality and standards.
• OpenMined: mentored 25 participants of the Beta Tech Bootcamp on
  Privacy-Preserving ML.
• Guavus: mentored two data-scientist interns on AI model integration.
• IDIR: trained three undergraduate CS students in Python and ML for their
  Co-op term.

Awards & Achievements

Certificates