Computer Vision for Automated Property Inspections - ON-1175
Project type: ResearchDesired discipline(s): Computer science, Mathematical Sciences, Operations research
Company: CMAI
Project Length: 6 months to 1 year
Preferred start date: 03/01/2026
Language requirement: English
Location(s): Toronto, ON, Canada
No. of positions: 1
Desired education level: Master'sPhD
Open to applicants registered at an institution outside of Canada: No
About the company:
CMAI is an advanced Artificual Intelligence (AI) technology company focused on transforming property management operations through artificial intelligence driven operation management and workflow automation. The firm not only unifies the entire end-to-end property managment workflows, but also leverages properties’ data, maintaince records and regulations to build a predictive serivce model to significantly drive property management efficiency, energy and reserouce saving, environment and social sustainbility, driving asset value growth.
Describe the project.:
Property management teams are currently overwhelmed by reactive maintenance requests. The current workflow is disjointed: tenants submit vague complaints with low-quality smartphone photos (e.g., "The sink is leaking"), while valuable historical data regarding that asset lies buried in unsearchable PDF maintenance logs. Without connecting these data points, managers cannot accurately judge the severity of an issue or predict if a minor leak is a precursor to a major system failure.
The main goal of this research project is to develop an Automated Maintenance Triage & Prediction Engine. Unlike standard ticketing software, this system utilizes Multimodal Machine Learning, combining two distinct data streams:
1. Visual Data: Images taken by tenants (via smartphone apps) and remote fixed cameras (in mechanical rooms or hallways).
2. Historical Data: Text-based maintenance logs, equipment age records, and previous repair invoices.
The objective is to train a Computer Vision (CV) model to automatically identify the defect in a photo (e.g., classifying "Active Water Intrusion" vs. "Cosmetic Paint Peeling") and cross-reference it with the building’s history. If a tenant snaps a photo of a ceiling stain, the AI should not just flag a "stain"; it should analyze historical records to predict: "High probability of HVAC pan overflow based on 3 similar incidents in this stack."
Ultimately, this project aims to shift maintenance from "Reactive/Manual" to "Predictive/Automated," reducing unnecessary truck rolls by 30% and enabling managers to catch systemic building failures before they escalate.
Required expertise/skills:
Applied Computer Vision & Data Fusion."
1. Technical Competencies (Hard Skills)
• Computer Vision (CV): Proficiency in Image Classification and Object Detection (e.g., ResNet, EfficientNet, YOLO) to identify defects in variable quality photos (poor lighting, blur).
• Multimodal Learning: Experience fusing different data types (Images + Text). For example, combining an image feature vector with a text embedding to make a prediction.
• Data Cleaning: Ability to handle "noisy" real-world data (e.g., filtering out irrelevant photos sent by tenants).
• Mobile/Web Integration: Basic knowledge of how to deploy models to a web backend (FastAPI/Flask) to process user uploads.
2. Research Focus Areas
• Automated Triage: Interest in "Recommender Systems" or decision support systems.
• User-Centric AI: Dealing with data generated by non-experts (tenants).
