Python

Overview Built for the Google Gemini AI Challenge 2026, this multimodal AI agent enables maintenance technicians to diagnose equipment issues by pointing a camera at a device and describing the problem through natural conversation. Problem Maintenance technicians often spend significant time diagnosing equipment issues — cross-referencing manuals, searching past incident reports, and consulting senior colleagues. This slows down repair times and increases downtime costs. Solution An AI agent that combines: ...

Overview Pennymize is a personal finance SaaS application designed for Canadian users. It combines traditional budgeting tools with AI-powered features for expense categorization, investment optimization, and personalized financial advice. Problem Existing personal finance apps don’t serve Canadian users well — they lack support for Canadian banks, tax structures (TFSA, RRSP), and the specific financial landscape. Most also lack intelligent automation for categorization and advice. AI Features Expense Categorization ML models automatically categorize transactions based on merchant names, amounts, and spending patterns — learning from user corrections over time. ...

Overview Designed classification models for Autism Spectrum Disorder (ASD) detection using Support Vector Machines and Convolutional Neural Networks, achieving 94% accuracy. This research was published in the International Journal of Information Technology and Computer Science (IJITCS). Publication Published in: International Journal of Information Technology and Computer Science (IJITCS), Vol.14, No.4 Approach Data preprocessing — Cleaned and prepared ASD screening datasets with feature engineering Model development — Implemented SVM and CNN architectures for binary classification Evaluation — Compared model performance using accuracy, precision, recall, and F1-score Visualization — Created Tableau dashboards to help healthcare professionals understand demographic patterns Tech Stack Python (Scikit-learn, TensorFlow) Support Vector Machines Convolutional Neural Networks Tableau Results 94% accuracy on ASD detection CNN outperformed traditional ML approaches on this dataset Visualization dashboards provided actionable insights for healthcare professionals Key Takeaway This project combined rigorous ML methodology with healthcare domain application, demonstrating how deep learning can assist in early screening for developmental conditions. ...

Overview A fully deployed machine learning system for house price prediction, built on AWS infrastructure with automated CI/CD pipelines. Architecture AWS Lambda — Serverless inference endpoint AWS EC2 — Model training and data processing AWS S3 — Model artifact and dataset storage GitHub Actions — Automated CI/CD pipeline for continuous deployment Features Dynamic feature testing — Users can adjust input features and get instant predictions Automated deployment — Push to main triggers model retraining and redeployment Scalable inference — Lambda-based endpoints scale automatically with demand Tech Stack Python (Scikit-learn, Pandas) AWS (Lambda, EC2, S3) GitHub Actions Docker Links Project Details Key Takeaway This project demonstrates end-to-end MLOps — from model development to cloud deployment with CI/CD. It covers the full lifecycle that production ML systems require. ...

As part of the Google Gemini AI Challenge 2026, I built a multimodal AI agent that helps maintenance technicians diagnose equipment issues. Here’s the story of what I built, why, and the key technical decisions along the way. The Problem Maintenance technicians at transit companies spend significant time diagnosing equipment failures. They flip through thick manuals, search through past incident reports, and call senior colleagues for advice. All while the equipment sits broken and operations are impacted. ...