Projects

Real problems, real solutions. Here's what I've built — at work, for clients, and on my own time.

Completed POCProfessional (WestJet)

Elsa — AI-Powered Document Intelligence

The Problem

10,000+ internal process and procedure documents were locked in proprietary software — inaccessible to the teams who needed them most. Finding the right procedure meant digging through folders, guessing filenames, or asking someone who might remember.

What I Built

Built an automated pipeline that exports documents from proprietary software as JSON, processes them to extract key variables and relationships, and feeds everything into a Snowflake Cortex AI agent. Validated the approach with both JSON and PDF exports — JSON proved more reliable for structured extraction. The agent handles natural language queries using RAG, so users can ask questions in plain English and get the right procedure back.

The Result

Completed proof of concept validating the full pipeline: document upload → processing → variable extraction → AI-ready format. Demonstrated that human agents could retrieve operational procedures through natural language queries — replacing hours of manual document hunting with a 30-second conversation.

Tech

PythonSnowflake CortexRAGJSON Processing
LivePersonal

FindMasajid.com — Prayer Time Aggregation Platform

The Problem

Muslim communities across Ontario had no single source for accurate prayer times. Every masjid posts them differently — some on websites, some in images, some buried in iframes — and they change seasonally. Families and individuals had to check each masjid individually.

What I Built

An automated data pipeline that scrapes, normalizes, and serves prayer times from 200+ sources daily. The system uses 10+ parsing methods to handle the wild variety of formats — iframe extraction, DOM parsing, regex, direct API calls — and falls back to OpenAI Vision API for layouts that resist structured extraction (think prayer times embedded in decorative images). The entire pipeline runs on n8n workflows with zero manual intervention. Data lands in Firebase Firestore and serves a clean, fast frontend.

The Result

Live and serving communities across Ontario with accurate, daily-updated prayer times. Planning expansion to additional provinces and automated detection of new masjid sources.

Tech

n8nJavaScriptFirebase FirestoreOpenAI Vision APIAutomated Data Pipelines
LiveProfessional (WestJet)

FlightClub — Employee Flight Search Platform

The Problem

14,000+ airline employees needed a way to search available flights for staff travel benefits. The existing process was manual, slow, and didn't scale.

What I Built

Led the project from zero — created the initial prototype and wireframes, validated the concept with business stakeholders, then partnered with vendor and internal teams to design data contracts and APIs. Led backend integration, testing, and technical sign-off for the full rollout.

The Result

30,000+ flight searches on launch day. Now a core internal tool used daily across the company. Became the model for how internal product ideas get validated and shipped.

Tech

Node.jsAzureAPI DesignVendor Integration

Built during my role at WestJet. Details kept general to respect confidentiality.

LiveProfessional (WestJet)

AI-Powered Workflow Automation — Intelligent Internal Placement

The Problem

When employees required modified duties (medical accommodations, temporary reassignments), the internal placement process took weeks. It was manual, involved multiple coordinators, and often stalled waiting on handoffs.

What I Built

An intelligent classification and routing system that automates the placement workflow end-to-end — from intake to matching to assignment. The system handles classification, eligibility checks, and coordinator notifications automatically.

The Result

Cycle time dropped from weeks to under 24 hours. Accuracy above 90%. The team went from chasing emails to reviewing recommendations.

Tech

Workflow AutomationClassification SystemsProcess Optimization