In partnership with WAT.ai

• LinkedIn: https://www.linkedin.com/in/linokee0423/
• Email: guitarkinglino@gmail.com
• GitHub: d-lino-kee
• Discord: dalinoman

• LinkedIn: https://www.linkedin.com/in/edsontakei/
• Email: etakei@my.yorku.ca
• GitHub: edsontakei
• Discord: luketonbeleu

FORTif.ai is an AI-driven companion that empowers seniors to live independently by merging proactive safety monitoring with tailored daily support. Using a computer-vision–powered Hazard Detection model, it continuously scans the home for potential risks—like spills, cluttered pathways, and tripping hazards—and offers clear, actionable recommendations to address them. At the same time, an intuitive AI chatbot engages users in friendly, proactive conversations, providing timely medication and appointment reminders, personalized wellness check-ins, and empathetic responses to questions or concerns. With built-in voice-to-text capabilities and real-time safety insights, FORTif.ai delivers a seamless, user-centric experience designed to enhance home safety, streamline everyday routines, and foster lasting independence for seniors.

1. 🤖 AI Chatbot (LLM): Design and deploy a robust conversational assistant

   • Develop reliable voice-to-text transcription with built-in quality control
   • Implement an action-oriented interface for booking appointments and sending medication reminders
   • Ensure LLM responses are accurate, relevant, and aligned with senior-friendly language
   • Integrate the chatbot with the back-end Hazard Detection system for contextual alerts
   • AI Topics: Speech Recognition, Prompt Engineering, LLM Evaluation, System Integration

2. ⚠️ Hazard Detection (OpenCV): Build and validate a vision-based model for home-hazard identification

   • Subgoal 1: Model development to detect obstacles along predefined walkways
   • Subgoal 2: Define performance metrics and risk thresholds (e.g., model accuracy and obstacle risk scoring)
   • Subgoal 3: Curate and preprocess high-quality training datasets (data cleansing, augmentation, transformation)
   • Quantify obstacle count and relative size for risk prioritization
   • Establish a 70:30 training–testing data split to evaluate generalization
   • AI Topics: Computer Vision, Object Detection, Data Preprocessing, Model Evaluation

3. 🔗 Integration: Deliver a seamless, end-to-end user experience

   • Push real-time hazard alerts and recommendations through the AI chatbot
   • Conduct end-to-end testing of voice, vision, and notification pipelines
   • Validate system performance in simulated home environments and user trials
   • AI Topics: Multimodal Integration, API Development, UX Design, Automated Testing

As the global population ages, many seniors face significant challenges in maintaining their independence while ensuring their safety. Common risks include falls, accidents at home due to environmental hazards, and difficulties in managing daily tasks, such as taking medications or attending appointments. These issues often lead to a decline in quality of life and can result in a need for constant caregiver assistance, which is not always feasible or sustainable.

The goal of this project is to address these challenges by providing seniors with an AI-driven assistant that promotes both safety and independence. By using computer vision for hazard detection, gait analysis for fall vulnerability assessment, and a user-friendly chatbot for daily reminders and check-ins, we hope to ensure that seniors can confidently navigate their living spaces while staying on top of their personal care routines. This project aims to reduce the risk of accidents, improve daily living, and empower seniors to maintain their autonomy in a safe, supportive environment.

Key epidemiological studies highlight the prevalence and consequences of falls among adults aged 65+, underscoring the need for proactive hazard detection in home environments. These reports inform our risk thresholds and help prioritize the most critical obstruction types.

• Nonfatal and Fatal Falls Among Adults Aged ≥65 Years
• Kakara, R., Bergen, G., Burns, E., & Stevens, M. (2023). CDC MMWR Report

Design guidelines for older adults emphasize clear visual cues, minimized clutter, and intuitive interfaces—principles we incorporate into both our physical hazard alerts and chatbot UX.

• Farage, M. A., Miller, K. W., Elsner, P., & Maibach, H. I. (2012). Global Journal of Health Science: Design Principles for Older Adults

Core CV methods—object detection, semantic segmentation, and AR overlays—enable real-time identification and visualization of hazards in complex home scenes.

• OpenCV Documentation
• OpenCV GitHub Repository

Practical tutorials and libraries streamline model prototyping, from image preprocessing in OpenCV to deploying YOLO for fast, accurate object detection.
OpenCV Resources

• OpenCV Tutorials (C++ & Python)
• OpenCV Python Tutorials
  YOLO Resources
• YOLO for Object Detection (Darknet)
• Aggarwal, A. (2024). YOLO Explained on Medium
• Kalra, K. (2023). YOLO Object Detection Overview

Classification benchmarks distinguish safe versus hazardous scenes, informing both the risk-scoring system and the chatbot’s contextual guidance.

• Custom Vision Service: Getting Started
• Image Classification on Papers With Code
• Viso.ai: Image Classification

TensorFlow and PyTorch underpin our model development, offering tools for CNN training, RL experiments, and seamless deployment pipelines.

• TensorFlow
• 5 Best Resources to Learn TensorFlow (Medium)
• PyTorch

Effective preprocessing and feature engineering rely on libraries like Pandas, NumPy, Matplotlib, and Scikit-learn—essential for cleaning CV datasets and visualizing model performance.

• Kaggle Discussion: Top Matplotlib Visualizations
• Machine Learning Plus: Master Plots in Python
• Awesome Scikit-Learn GitHub

Gait analysis research identifies movement patterns and common abnormalities in seniors, informing our hazard-scoring algorithms and risk thresholds.

• Gait Patterns in Older Adults (PMC)
• Gait Variability & Fall Risk (PMC)
• Rehabilitation Biomechanics (Nature)

CNN architectures power feature extraction and classification in our hazard detection pipeline, enabling the model to learn visual patterns of risk versus safety.

• IBM: Convolutional Neural Networks

RL techniques guide the development of a dynamic scoring system, where an agent learns to assign risk levels based on environmental feedback and obstruction severity.

• Getting Started with Reinforcement Learning (Medium)
• UBC CPSC533: Reinforcement Learning

We leverage a centralized Notion workspace to plan, track, and collaborate on every aspect of FORTif.ai. Our Notion hub serves as the single source of truth for:

• Roadmaps & Timelines: Interactive Gantt charts and calendar views keep milestones visible.
• Task Boards: Kanban-style boards for both the AI Chatbot and Hazard Detection subteams.
• Documentation: Live specs, meeting notes, design guidelines, and API references.
• Knowledge Base: Central repository for background research, UX guidelines, and stakeholder feedback.

Access our Notion hub here:
🔗 FORTif.ai Workspace Hub

Role: Hazard Detection Model Development Lead

Lino Kee is an undergraduate student at the University of Waterloo studying Honours Management Engineering. He brings academic and practical co-op experience in project management, data analysis, automation development, quality assurance, business platform management, and software development. Lino leads the development of the computer-vision–driven Hazard Detection model and oversees the overall architecture and delivery of the FORTif.ai tool. He’s happy to answer questions about project scope, timeline, and the long-term vision for FORTif.ai.

Role: AI Chatbot Lead

Role: AI Chatbot Lead

Edson Takei is un undergraduate student at York University studying Software Engineering in the Big Data stream. Edson has academic and applied experience in data analysis, genAI, bioinformatics and digital public health having also pursued research opportunities at the University of Waterloo and McGill University. His work includes leveraging large language models (LLMs) in healthcare to automate manual processes, as well as evaluating the effectiveness of LLMs in public health from both technical and policy-oriented perspectives. Edson leads the development of the AI chatbot subteam of FORTif.ai and oversees the overall architecture and delivery of the FORTif.ai tool. He’s happy to answer questions about project scope, timeline, and the long-term vision for FORTif.ai.

• Dhruv Sharma (@dhruvsharma11) Hello, my name is Dhruv Sharma and I'm a 4th year Management Engineering student! My main interests include playing and watching sports such as basketball and cricket. I also really enjoy playing video games. My main role on the team is to help with the fall detection model!

• Naomi Eshetu (@naomieshetu) Hi my name is Naomi and I'm a 3rd year Biomedical Engineering student! I enjoy reading, roller skating, and watching true crime. My role on the team is to do hazard detection and biomechanics research.

• Marco Kee (@Keezyyy) Hello! My name is Marco and I’m a 2nd year Health Sciences student. I love watching sports (especially football), going to the gym, and watching a new movie with friends. I’m involved with the hazard detection and biomechanics research for this project!

• Adnan Habib (@adnxnhabib) Hey! I'm Adnan and I'm a 4th year CS/BBA student. I love video games, movies, going to the gym and chilling. I'm part of the hazard detection team helping to build our model!

• Owen Kim (@Owenkim2006)
  Hi, I'm Owen and I'm a first year BME student. I am interested in learning about new developments in ML/AI, and I love playing many sports such as basketball, volleyball and ultimate frisbee. I will be working on the hazard detection and biomechanics sub team for this project.

• Sania Banga (@Saniabanga)
  Hi! My name is Sania and I’m a 4th year student majoring in CS. I love reading, hiking and watching horror movies. My role on the team is to help with the fall and hazard detection!

• Meghana Yarlagadda (@yarlas99)
  Hello I am Meghana, 3rd year Statistics and Computational Mathematics student. I will be working on the Fall and hazard detection team.

• Anahad Dhaliwal (@Anahadd)
  I'm Anahad, first year CE student. I love coding and watching sports. I'm very interested in LLMs.

• Larris Xie (@Profilist)
  Hi, I'm Larris, a 1st year CS student at the University of Waterloo. I enjoy playing the piano and basketball in my free time. I'm currently working on LLM agents for FORTif.ai!

• Akil Giri (@akilgiri) Hi, I’m Akil, a 3rd year Computer Engineering student at the University of Waterloo. I enjoy playing video games and going on walks in my free time. I’m currently working on AI Chatbot team on FORTif.ai

• Lucas Khan (@1-mbps) Hi, I'm Lucas and I'm a 3rd year Computational Math student. I'm working on the chatbot for FORTIF.ai. In my free time I enjoy running, watching movies, and watching football.

Our stakeholders include the family caregivers and seniors whose real-world needs drive FORTif.ai’s design:

• Family Caregivers:
  Lino and Edson both support elderly grandparents who face memory loss, mobility challenges, and require ongoing assistance for daily tasks. Their firsthand insights into appointment scheduling, medication reminders, and home-safety concerns inform every feature decision.

• Professional & Informal Caregivers:
  Nurses, aides, and family members responsible for elderly loved ones will validate that FORTif.ai remains intuitive, engaging, and low-friction, reducing their own cognitive load while empowering them to monitor and support seniors more effectively.

• Senior End-Users:
  Seniors themselves—often living with multiple health issues—will provide critical feedback on conversational tone, alert frequency, and interface simplicity. Their lived experience ensures the assistant feels like a trusted companion rather than another complex tool.

• Subject-Matter Contributors:
  While they may not code or draft technical documentation, seniors and caregivers bring invaluable domain knowledge—from gait challenges in the home to typical medication schedules—that shapes our hazard thresholds, reminder logic, and overall user experience.