Maryam Gohargani
Toronto, ON · mgohargani@gmail.com
Applying for graduate study in trustworthy ML, starting Fall 2027. Also open to independent research and collaboration, and always looking to learn something new.
Education
University of TorontoSeptember 2019 – April 2022
Honours Bachelor of Science, Computer Science, Minor in Mathematics
GPA 4.0 / 4.0 · 95% average
Dean's List Scholar (2020–2022) · Honour Roll (2020–2022) · Entrance Scholarship (2019)
Selected coursework:
Research interests
- Privacy and security of machine-learning systems
- Transparency and evaluability of intelligent systems
- AI misuse and unintended consequences
- Human–AI interaction and trust
- AI in education and learning
AI safety
Label-Only Membership InferenceJuly 2026 – Present
Paper reproduction — PyTorch, CIFAR-10, scikit-learn
- Reproducing Choquette-Choo, Tramèr, Carlini & Papernot (ICML 2021), training the target models and implementing the attacks
- Gap-attack baseline reproduced exactly: membership-inference accuracy 0.719, matching the theoretical ½ + (acctrain − acctest)/2
- Data-augmentation attack (47 label-only queries per point) reaches 0.759, beating the gap baseline, and a naïve count-threshold variant scores 0.685, showing the gain comes from weighting the queries rather than from augmentation robustness alone
- Documented throughout as a public write-up series, including a from-scratch derivation of backpropagation (read the series)
Model-Written EvaluationsJuly 2026 – Present
Independent research project — Python, LLM APIs (Anthropic, Gemini), Ollama
- Reproducing Perez et al. (2022) model-written evaluations against current frontier models; built the evaluation harness from scratch (multi-provider, with offline re-analysis)
- Established each dataset's label orientation directly from the data — which contradicts the paper's stated convention — before interpreting any results
- Found survival-instinct and power-seeking behaviours decrease with model capability across Claude Haiku, Sonnet and Opus, partly reversing the paper's original scaling trend
Trajectory LabsJune 2026 – Present
Member — Toronto's in-person AI-safety community
Attending regularly to keep current with alignment research and to work alongside others building in the field; where much of my present research direction took shape.
Teaching experience
University of TorontoSeptember 2020 – April 2022
Teaching Assistant — Computer Science & Mathematics
15 course sections across Computer Science and Mathematics:
- Designed and delivered tutorials and practicals for groups of up to 90 students; held regular office hours
- Managed discussion boards for courses with 800+ enrolment; graded assignments, tests and exams for 500+ students
- Devised and proofread course materials, designed problem sets and exams, and determined grading rubrics
- Graded proof-based mathematics assignments with detailed written feedback
- Supported student learning through the COVID-19 pandemic, adapting to fully remote instruction
Pre-university (gap year)2018 – 2019
ESL Teacher & Informatics Instructor
- Certified ESL teacher; taught English language courses independently
- Designed and taught an algorithms and programming curriculum for high school girls preparing for the International Olympiad in Informatics (IOI)
Industry experience
Software EngineerApril 2022 – July 2023
Avenu Insights & Analytics (formerly Harmari by LTAS Technologies)
- Developed backend software in Python, Java, PHP and MySQL for municipal and regulatory technology products
- Designed data processing pipelines to parse and transform KML/XML and geospatial datasets, saving 5+ hours/week for client-facing teams
- Built automated QA tooling using Python and Matplotlib to validate accuracy of data scraped from external sources
- Implemented and enhanced cross-checking algorithms to match short-term rental data with licensing records (60% viable match rate)
- Refactored and redesigned MySQL database schemas, improving performance and eliminating 1,000+ lines of duplicate code
- Migrated legacy Java and PHP/Laravel codebases to support updated data models
- Deployed and maintained applications on remote customer servers under strict jurisdictional data-security requirements
- Researched, proposed and evaluated technical solutions using trade-off analysis; presented recommendations to leadership
- Led the team's migration from SVN to Git, establishing workflows and training colleagues on version-control practices
- Debugged production issues and improved system reliability for customer-facing and internal users
- Worked directly with clients to clarify requirements, troubleshoot issues and iterate on solutions
Backend Software DeveloperJune 2021 – November 2021
University of Toronto Mississauga, Dept. of Mathematical and Computational Sciences (volunteer)
- Recruited by the department to help build processes and tools supporting student academic success
- Worked in a team of five to revitalise the Computer Science Student Community website
Technical skills
Machine learning: PyTorch, torchvision, scikit-learn, NumPy, training CNNs from scratch, GPU training on Google Colab
ML security & evaluation: membership-inference attacks, model-behaviour evaluation harnesses, LLM APIs (Anthropic, Gemini), local models via Ollama
Languages: Python, Java, C/C++, JavaScript/TypeScript, PHP/Laravel, SQL, Haskell, Racket, MIPS Assembly, R, Bash
Data & geospatial: Pandas, GeoPandas, Shapely, NumPy, Matplotlib
Tools & infrastructure: Git/GitHub, Linux/Unix, Docker, Postman, SVN, CI-friendly workflows, customer-deployed environments
Backend & APIs: FastAPI, RESTful APIs, Pydantic, Node.js, Express, NestJS, Spring, service-oriented architecture
Databases: PostgreSQL, MySQL, MongoDB, Neo4j
Frontend: React, Next.js, Astro, Vue, Nuxt, HTML/CSS, Storybook, Figma
Projects
Invoice & Payments SystemFebruary 2026
FastAPI, React, TypeScript, PostgreSQL
- Built full-stack invoice & payment app with CRUD, invoice lifecycle (DRAFT → PENDING → PAID/VOID) and partial payment support
- Implemented row-level locking to prevent overpayment under concurrency; enforced business rules with Pydantic validation
- Delivered React SPA with list/detail/create flows, filters (status, customer, date range), and 50+ pytest tests
Toronto Public Resource Density APIJanuary 2026
Python, FastAPI, GeoPandas, Pandas, Shapely
- Designed a FastAPI microservice to ingest municipal GeoJSON datasets and expose analytics through RESTful APIs
- Built service-oriented architecture with separate ingestion, processing and API layers; engineered geospatial pipelines with CRS transformations and spatial joins
-
Implemented a
/densityendpoint computing and serving neighbourhood-level resource-density metrics as structured JSON - Applied backend best practices: Pydantic schema validation, modular routing and separation of concerns
Baobab2022
MERN, TypeScript, REST APIs, three-tier architecture
- Built a community-driven e-learning platform supporting multiple user roles, team formation, posts and mentor feedback workflows
- Designed RESTful backend APIs with TypeScript and Express; modelled MongoDB schemas for users, teams, posts and submissions
- Led a team of seven developers, delegating tasks and running agile stand-ups
Community and leadership
VP of External AffairsMay 2021 – May 2022
Mathematical & Computational Sciences Society, University of Toronto Mississauga
- Elected student leader responsible for academic and non-academic events for 1,000+ students
- Led organisation and delivery of UTM's largest Hackathon (200+ participants)
- Conducted technical interviews to recruit student developers for the society's website team
ACM-ICPC Team Member2020 – 2021
University of Toronto
Mentor, Women in Science & Computing2021
University of Toronto Mississauga
Mentor, International Student Mentorship Program2020
University of Toronto Mississauga
Study Group Leader, Peer Facilitated Study Groups2020
University of Toronto Mississauga
Event Volunteer, Math in Motion… Girls in Gear!2019
University of Toronto Scarborough