mahdi alam.

SOFTWARE DEVELOPER · TORONTO, ON

Good software.
Grounded in
good thinking.

I’m Mahdi. I build Python systems and machine learning pipelines, with a research background that keeps me asking: does it actually work better?

Explore my work GitHub
FROM COMPLEXITY TO CLARITYFIG. 01
A coarse grid transforms into a refined grid, representing Mahdi’s coarse-to-fine research approachCOARSEFINE
Break it down. Build it up.MA / 01
Built on a foundation in

Computer Science / AI Research / Reproducible Engineering

01 / SELECTED WORK

A few problems
I’ve worked through.

From medical images to search results.
Different problems, the same care for the details.

RESEARCH SYSTEMS 01
01 / INPUT
02 / ALIGN
COARSE → FINE → EVALUATEConcept illustration

MACHINE LEARNING / UNIVERSITY RESEARCH

Better alignment.
Same architecture.

Improving self-supervised medical image registration through a two-phase training strategy: multi-resolution curriculum learning and adaptive optimization.

+1.7 points

Up to 1.7-point improvement in Dice Similarity Coefficient across public medical-image datasets.

PythonSelf-supervised learningOptimization
Project documentation on GitHub
Read the engineering breakdown

The problem

Align medical images while evaluating both overlap accuracy and deformation quality.

The approach

Extended the On-the-Fly Guidance framework with a coarse-to-fine training curriculum and adaptive optimization, without changing model architecture or inference pathways.

The evaluation

Ran reproducible experiments, ablation studies, and benchmark comparisons on LPBA40, IXI, and OASIS. Used Jacobian determinant analysis to evaluate deformation quality.

The related manuscript was listed as in preparation in the supplied resume (2025).

INFORMATION RETRIEVAL 02
Find the most relevant document
01
Term relevance
BM25
02
Document similarity
TF–IDF
03
Classification baselines
SVM
QUERY → RANK → COMPAREConcept illustration

PYTHON / NLP · OCT–NOV 2022

Finding signal in text.

A search engine project comparing retrieval methods and classification baselines to understand what makes a result relevant.

TF–IDFBM25Naïve BayesLinear SVM
Project documentation on GitHub
Read the engineering breakdown

Implementation

Implemented TF–IDF and BM25 retrieval in Python, then trained Naïve Bayes and linear SVM baselines.

Evaluation

Compared retrieval effectiveness across models. The project brings together ranking, text representation, and quantitative comparison.

COMPUTER VISION 03
EXTRACT → CLUSTER → VALIDATEConcept illustration

PYTHON / SCIKIT-LEARN · SEP–DEC 2023

Making colour patterns visible.

A colour clustering pipeline that extracts dominant distributions, finds recurring palettes, and evaluates clustering stability.

PythonScikit-learnK-means
Project documentation on GitHub
Read the engineering breakdown

Implementation

Extracted dominant colour distributions and applied K-means clustering to identify recurring palettes.

Evaluation

Evaluated clustering stability, connecting the visual output to the behaviour of the underlying model.

Explore my GitHub profile

02 / BACKGROUND

A researcher’s care.
A developer’s craft.

My work sits at the intersection of software and machine learning. I’m interested in the systems around a model as much as the model itself: how experiments run, how results are compared, and how improvements can be reproduced.

I bring a computer science foundation, hands-on Python experience, and a habit of testing ideas against evidence.

Read my full resume

EDUCATION

MSc · Artificial Intelligence

Research-based thesis
Toronto Metropolitan University

2023 — Expected Dec 2026

BSc (Hons) · Computer Science

Toronto Metropolitan University

2019 — 2023

TOOLS I WORK WITH

Languages
Python, C++, C
ML & data
PyTorch, TensorFlow, NumPy, Pandas, SciPy, OpenCV, Scikit-learn
Engineering
Git, Docker, Linux, reproducible experimentation

03 / WHAT’S NEXT

Let’s build
something useful.

I’m looking for software and machine learning internships and early-career opportunities where I can contribute, learn, and solve worthwhile problems.

LinkedIn ↗GitHub ↗Resume ↗