Parham Kazemi

Print

PARHAM KAZEMI

Vancouver, BC pkazemi3@gmail.com +1 (604) 727-1834
parham-k.github.io github.com/parham-k linkedin.com/in/p-kazemi

Machine Learning Engineer and Software Developer specializing in deep learning (Python/PyTorch) and high-performance computing (C++). Experienced in integrating large-scale computational models into highly optimized, production-grade software. Proven track record of training CNN/Transformer models on GPU clusters and designing efficient bioinformatics algorithms for 100GB+ to terabyte-scale datasets (4 first-author papers, 1 patent).

CORE SKILLS

Programming Languages: C++, Python, Java, SQL, Bash

ML & AI: PyTorch/libtorch, HuggingFace, RL, Self-Supervised Learning, Quantization, ONNX

Software Engineering: Django, REST APIs, PostgreSQL, Docker, Nginx, Git, CI/CD, CMake, pybind11

High-Performance Computing: OpenMP, SIMD Vectorization, Memory Optimization, CUDA, SLURM, Valgrind

Domain & Data Specialties: NLP, Bioinformatics, Unstructured Data, Sequence Analysis, Signal Processing

EXPERIENCE

Graduate Research Assistant

BC Cancer Research Institute (September 2021 - Present)

  • Architect and manage the full software development lifecycle of 4 open-source genomic libraries and pipelines.
  • Develop high-performance C++ pipelines with multi-threading (OpenMP) and memory optimization to accelerate sequence analysis for 100GB+ datasets.
  • Ship Python bindings via pybind11 and automated CI/CD with package release pipelines (Bioconda).
  • Train self-supervised CNN and Transformer models on terabyte-scale signal datasets with Flow Matching and Gromov-Wasserstein Optimal Transport, using FlashAttention, CUDA, and SLURM-managed GPU clusters.

Backend Developer and System Administrator

University of Isfahan (September 2018 - September 2021)

  • Built an Alumni Social platform in Django with PostgreSQL serving 100,000+ alumni, including a credential verification engine with institutional syncing.
  • Administered the production stack on Linux (Ubuntu) using Nginx and uWSGI on the university’s networking platform that remains in active production today.
  • Developed a ticketing system integrating SMS APIs and payment gateways for event registration.

Course Instructor and Teaching Assistant

University of Isfahan (September 2016 - June 2020)

  • Instructed Python and Django courses for the ACM Student Chapter (~20 students per cohort).
  • Assisted across 12 sections of 9 courses (~30 students each) including AI, Algorithm Design, Data Structures, and Advanced Programming in Java and C++.
  • Built an automated grading platform from scratch using Python, socket programming, and Tkinter, with C++ and Java interfaces for student AI project submissions.

EDUCATION

PhD in Bioinformatics

University of British Columbia (September 2021 - Expected 2026)

  • Thesis: Computational Representations for Nanopore Basecalling and Assembly Polishing
  • Supervisor: Dr. Inanc Birol

MSc in Computer Engineering

University of Isfahan (September 2019 - June 2021)

  • Thesis: Deep Reinforcement Learning for Training Intelligent Agents in Natural Language Environments
  • GPA: 18.42/20 (highest in cohort)

BSc in Computer Engineering

University of Isfahan (September 2015 - June 2019)

  • Thesis: Predicting Persian Twitter Users’ MBTI Personality Using Text Mining Methods
  • GPA: 18.25/20 (highest in cohort)
  • Finalist, National Computer Engineering Olympiad (2019); ACM-ICPC West Asia Regional contestant (2016, 2017); RoboCup IranOpen 2D Soccer Simulation competitor (2017)

PROJECTS AND PUBLICATIONS

Myrid: Self-Supervised Deep Learning for Raw Signal Processing

Provisional U.S. patent filed in 2026

  • Architected a self-supervised sequence model trained directly on raw nanopore signals without labeled data, eliminating chemistry-dependent model retraining.
  • Presented model benchmarks and architecture as a poster at RECOMB 2026 (Thessaloniki, Greece).

AIEdit: Alignment-Free ML-Based Genome Assembly Correction

GitHub: BirolLab/AIEdit - PLOS CB: 10.1371/journal.pcbi.1014245

  • Engineered deep learning sequence models in PyTorch and compiled them directly into C++ using TorchScript and pybind11 for low-latency inference.
  • Achieved 58% error reduction vs. 21% for prior methods, dropping run times from days to 2.7 hours using 3× less memory on human-scale data.

ntStat: Toolkit for Statistical Analysis of K-mer Frequency and Depth

GitHub: BirolLab/ntStat - PLOS CB: 10.1371/journal.pcbi.1014158

  • Tracks k-mer/n-gram count and depth de novo using Bloom filter data structures, maintaining 99.5-99.9% accuracy at reduced memory usage.
  • Components written in Python and C++, integrated together with pybind11.

ntHash2: High-Throughput Rolling Hash Algorithm

GitHub: BirolLab/ntHash - Bioinformatics: 10.1093/bioinformatics/btac564

  • Up to 2.1× faster than ntHash 1 and 3.8× faster than conventional algorithms at hashing spaced seeds.
  • Written in C++, Python wrappers provided by SWIG.

Fuzzy Word Sense Induction and Disambiguation

IEEE Transactions on Fuzzy Systems: 10.1109/tfuzz.2021.3133905

  • Uses fuzzy clustering on word embeddings to disambiguate meanings based on surrounding context.
  • Achieves F-scores of 0.58 and 0.62 on SemEval 2010 and 2013, respectively.

SELECTED TALKS

VOLUNTEER EXPERIENCE

Student Mentor & Conference Adjudicator — UBC (2024 - 2025)

  • Mentored 5 students through UBC’s Undergraduate Research Opportunities program.
  • Adjudicated undergraduate research posters at UBC’s Multidisciplinary Undergraduate Research Conference.

Volunteer Organizer — Vancouver Bioinformatics User Group, VanBUG (2023 - 2024)

  • Coordinated community engagement for seminars by local and international bioinformatics researchers.