Ali Lesani

Ali Lesani

Ali Lesani

AI and Computer Vision Researcher

PhD Student at CViSS Lab

Hi, I'm Ali!

I am an Artificial Intelligence researcher and PhD student at the University of Waterloo, researching 3D scene understanding and visual reasoning through vision-language models. I have hands-on experience developing AI systems, collaborating with academic and industry partners such as Rogers Communications and RBC Royal Bank of Canada. Proficient in Python and leading ML/AI frameworks (PyTorch, Hugging Face, scikit-learn, OpenCV), my expertise spans building and deploying end-to-end deep learning pipelines. I also have strong experience in Linux environments with Bash scripting, and a proven track record of delivering impactful AI solutions through collaborative research and engineering efforts.

Experience

Machine Learning Engineer

University of Waterloo

2024 - Present

AI & ML Researcher (PhD)

CVISS Lab, University of Waterloo

2023 - Present

View All Experience →

Education

PhD, Engineering

University of Waterloo

2023 - Present

M.Sc., Engineering

Sharif University of Technology

2019-2022

View All Education →

Technical Skills & Tools

Python

Python

C++

C++

C

C

Pytorch

PyTorch

HuggingFace

HuggingFace

scikit-learn

Scikit-learn

OpenCV

OpenCV

Docker

Docker

Bash

Bash

Git

Git

Linux

Linux

SQL

MySQL

Pandas

Pandas

Qt

Qt

Experiences

Machine Learning Engineer 2024 - Present

University of Waterloo - Office of Associate Dean, Teaching & Student Experience

  • Contributed to the development an NLP-based AI system to classify and filter open-ended student feedback using embedding models (e.g., BERT, Sentence Transformers).
  • Applied K-Nearest Neighbors (KNN) for supervised text classification and filtering based on predefined content categories.
  • Developed preprocessing and filtering pipelines in Python for text cleaning, feature extraction, and batch processing.

AI & Machine Learning Researcher (PhD Student) 2023 - Present

CVISS Lab, University of Waterloo

  • Computer Vision-Based Inspection of Cell Towers - in partnership with Rogers Communications
    • Designed a vision pipeline combining saliency detection, depth estimation, object detection, and Segment Anything (SAM) for multi-view cell tower component segmentation.
    • Investigated natural language-guided 3D scene reasoning and editing with text-to-geometry workflows.
  • Vision-Language AI for Automated Energy Audits - in partnership with RBC Royal Bank
    • Built a monocular vision-based measurement tool for estimating indoor object dimensions using depth maps and object detection.
    • Contributed to the development of a knowledge-augmented chatbot for energy audits using Retrieval-Augmented Generation (RAG).
    • Contributed to the development of a multi-agent system powered by LLMs (e.g., GPT, Gemini) to automate energy audit steps and reporting.
  • Graph Attention Network for Spatial Correlation Analysis
    • Contributed to the development of a Graph Attention Network-based architecture that fuses high-fidelity visual embeddings from UAV imagery with geospatial and structural metadata.
    • Outperformed CNN baselines by leveraging spatial attention mechanisms for building-level damage estimation.

Software Developer & Graduate Researcher 2020-2023

INSURER Lab, Sharif University of Technology

  • Worked as a developer on Rtx, a C++ and Qt-based simulation software for probabilistic modeling, reliability, and resilience analysis of infrastructure systems.
  • Implemented agent-based models to simulate disaster impacts and interdependent infrastructure behavior.
  • Developed modules for assessing community-scale risk and resilience across varying levels of model refinement.
  • Contributed to probabilistic simulation workflows for evaluating post-disaster recovery and system performance.

Selected Projects

Intelligent Decision-Making System: Deep Learning-Based Satellite Imagery Analysis for Probabilistic Inference via Bayesian Networks

  • Developed a deep learning pipeline that segments building footprints from satellite imagery using U-Net and classifies structural damage based on ResNet feature representations.
  • Integrated vision-based damage assessments into a Bayesian Network to perform probabilistic inference of community-level impacts.
  • Designed a modular pipeline enabling dynamic updates to inference as new imagery or evidence becomes available.

Education

Doctor of Philosophy, Engineering

University of Waterloo

2023 - Present | GPA: 99.67/100 (4/4)

Master of Science, Engineering

Sharif University of Technology

2019-2022 | GPA: 17.83/20 (3.87/4)

Bachelor of Science, Engineering

Sahand University of Technology

2015-2019 | GPA: 19.12/20 (3.96/4)

Publications

  • Lesani, A., Yeum, Ch. (In Preparation): "MonoWindow: Monocular 3D Window Measurement in Indoor Environments."
  • Lesani, A., Yeum, Ch. (In Preparation): "SDG-SAM: Salient-Depth-Grounded SAM for Robust Cell Tower Component Segmentation in Complex Visual Scenes."
  • Liu, Sh., Han, K., Lesani, A., Yeum, Ch., Watt, G., Lee, S. (2025): "A knowledge-augmented GPT for efficient data collection in home energy audits." Proceedings of the 6th International Conference on Building Energy and Environment (COBEE).
  • Lesani, A., Ahmadi, A., Costa, R. (2024): "Simulating Housing Recovery Challenges Due to Post-disaster Repair Cost Increases." Proceedings of the 18th World Conference on Earthquake Engineering (WCEE). [Link]