$ ls -a ./projects

Adio

Voice AI Repair Companion

Adio is a voice-first repair companion for home appliances and basic car fixes. It executes one procedural step at a time with confirmation-gated progression and interruption commands. Built for the AI Agents Waterloo Voice Hackathon. Features parallel STT race (smallest.ai Pulse + OpenAI Realtime), smallest.ai Waves streaming TTS, Supabase Postgres + pgvector RAG, and a YouTube Guide Mode that compiles video transcripts into step-by-step repair procedures.

Computare

Open-Source Personal Finance Platform for Canadians

An open-source personal finance platform that extracts transactions from Canadian bank statements (Scotiabank PDF, Wealthsimple CSV, Amex), categorizes them with a 3-tier AI pipeline (rules → merchant cache → GPT-4o-mini), detects recurring subscriptions, and stores everything in PostgreSQL with Row Level Security. Features a Next.js dashboard, FastAPI backend, and 9 materialized views for analytics.

Microsoft Workspace Skill

Microsoft Graph Integration for Agent Workflows

A Hermes Agent skill that connects Outlook email, calendar, contacts, and profile data through Microsoft Graph. It supports OAuth setup, token refresh, email search and sending, calendar scheduling, free/busy lookup, Teams meeting creation, and safe preview-based email sending for agent-assisted productivity workflows.

kag

Kaggle Competition Workspace TUI

A Textual-based terminal app that turns Kaggle competition setup into a repeatable workflow. It provides a searchable competition picker, scaffolds notebooks, notes, data folders, virtual environments, and git repositories, checks Kaggle access, and enriches project notes from competition pages.

Municipalities Budget Scraper

AI-Powered Municipal Financial Report Discovery

An intelligent web scraper built in support of @BuildCanada that automatically discovers and downloads annual audited financial reports from Canadian municipality websites. Features AI-powered document discovery via OpenAI, Firecrawl API for web crawling, Playwright browser automation, and a 4-strategy PDF extraction pipeline. Processes all 3,696 Canadian municipalities with URL caching for 10x faster subsequent runs.

Municipal Waste Schedule ETL

Public-Sector Waste Collection Data Pipeline

A Python ETL pipeline that extracts municipal waste collection schedules from ReCollect-powered services, augments Peel recycling data from Circular Materials, and normalizes nested event JSON into analytics-ready schedule outputs. The pipeline supports multiple Ontario municipalities, audit-friendly raw response storage, retry-aware API clients, and staged extraction, transformation, and merge scripts.

Public Jobs Scraper

Cross-Jurisdiction Government Job Research Pipeline

A research-grade scraping system for technical government job postings across Canada, Australia, and the United Kingdom. It monitors 14 jurisdictions, applies fuzzy title matching to reduce false positives, standardizes job details for comparative public-administration analysis, supports parallel Playwright collection, and includes Supabase upload workflows with GPT-assisted extraction for inconsistent pages.

LinkedIn Job Scraper

Intelligent Job Data Extraction with LLM Enhancement

A comprehensive LinkedIn job scraping tool that extracts job postings with intelligent parsing, LLM-enhanced skill extraction, and MongoDB storage. Features regex/NLP-based extraction combined with OpenAI GPT for skills analysis, a webhook server with Chrome browser extension for one-click job saving, MongoDB with auto-deduplication, and intelligent parsing of employment type, salary, benefits, and location.

LinkedIn Company Scraper

Company Enrichment for Job Search Data Pipelines

A Python scraper that extracts structured company data from LinkedIn pages, including company IDs, industry, size, headquarters, specialties, websites, and descriptions. It supports batch processing, rate limiting, optional Gemini URL verification, Supabase storage, and company ID extraction for downstream job search and business intelligence workflows.

Deep Learning CNN Regularization

CIFAR-10 Image Classification (UWaterloo STAT 940)

A comprehensive exploration of CNN architectures and regularization techniques for CIFAR-10 image classification. Systematically compares 6 model architectures—from a 2-layer baseline to a fully optimized model with batch normalization, dropout, and data augmentation.

House Price Prediction

End-to-End ML Pipeline for Residential Home Sales

An end-to-end machine learning pipeline for predicting residential home sale prices using linear regression. Covers 2006–2010 housing sales including the 2008 financial crisis. Final model achieves R² = 0.89 with ~$26,000 average prediction error.

Driving Performance

Analyzing Canada's Automotive Stocks and Export Trends

This interactive dashboard compares the cumulative returns of leading Canadian automotive companies against the TSX Index while visualizing provincial auto export performance. Built in R Shiny using tidyquant, tidyverse, and Statistics Canada data.

Canadian Census Data Explorer

An R Shiny app that lets users select Census year (2011/2016/2021), pick top Canadian cities, choose geography (CT/DA/CSD), preview results, and download a ZIP (CSV + metadata). Powered by cancensus + Statistics Canada vectors.

Predicting the Future of Electric Vehicles

A Study on BEV Car Manufacturer Stock Returns

This paper explores predictors of stock market returns for Battery Electric Vehicle (BEV) startups using GARCH and VAR models to forecast volatility and returns.

Understanding Public Opinion

Text Similarity and Sentiment Analysis

This project analyzes the relationship between an article and its comments using text similarity and sentiment analysis techniques. Insights are visualized with Gephi to highlight key themes and connections.