Software that ships safely, built by an engineer who has managed risk.

Non-functional testing: Dolibarr performance, usability and reliability

Problem. Could an open-source business application handle many users at once, stay usable, and recover when its servers failed?

What I did. Team project for Non-Functional Testing (INFO-6055). Wrote Apache JMeter tests for two user flows (login, add member, logout, and login, job application, logout) at 50 and 75 virtual users, then stress runs of up to 100 users. Audited pages with Chrome Lighthouse, and stopped Apache and then MySQL during a load run to compare recovery.

Result. Throughput reached 12 to 15 requests per second, above the 9 per second target, but login was the weak spot: at 50 users 90% of logins took up to 0.69 seconds against a 0.6 second target, and at 75 users in the job-application flow they took up to 2.4 seconds, with some over 12 seconds. Errors rose above the 1% limit on the add-member flow past about 12 requests per second, and at 100 users every step of the job-application flow failed at high rates. Lighthouse scored accessibility at 78 and 79 and SEO at 50 and 54. An Apache outage caused more errors than a MySQL outage, but the MySQL outage made the run take longer to finish.

Load testing a web application with JMeter: INFO-6055 final exam

Problem. How many users can a Tomcat-hosted shopping website serve before it slows down?

What I did. Individual practical for Non-Functional Testing (INFO-6055). Deployed the site on Tomcat through XAMPP and built a JMeter plan that visits the home, products, about us and contact us pages, with response assertions to confirm each page opens, a duration assertion of 2000 ms or less, and contact form data fed from a CSV file. Ran it at 1, 10, 50, 100, 150 and 200 users and recorded the Summary and Aggregate reports.

Result. No errors at any load. Throughput peaked at 81.4 requests per second at 50 users and fell to 26.5 at 200, while average response time stayed between 3 and 6 ms and the 99th percentile between 11 and 22 ms. I reported 50 users as the peak and recommended tuning the Tomcat thread pool.

Capstone: Money Manager EX application testing

Problem. Did a personal finance app work correctly on its web, desktop and Android versions, and did data move between them?

What I did. Five-person capstone team (Codebug Busters, INFO-6068). Wrote a test strategy and Excel test cases covering smoke, functional, invalid-input, boundary, usability, integration, load and compatibility testing, used Selenium for automation, load-tested the website with JMeter at 100 virtual users, and tracked defects in Excel.

Result. Executed test cases passed at 90% for bank accounts, 87% for budgeting and 80% for assets, with login, transactions and reports at 100%. Integration testing found a real defect: transactions made on the web appeared on the desktop app, but desktop transactions did not appear on the web. The team recommended a root-cause analysis and a fix to the platform synchronization.

The JMeter load test of the website ran twice with 100 users, once ramped over 10 seconds (average 39 ms, 10.2 requests per second) and once over 60 seconds (average 36 ms), with no errors in either run.

Selenium WebDriver test framework for OrangeHRM

Problem. How can you test an HR system with a modern, dynamic interface in a way that is repeatable and leaves evidence of every run?

What I did. Team project for Automated Test Tools (INFO-6105). Built a Java, Maven and TestNG framework with 13 automated tests covering login and logout, adding, searching and deleting employees, creating employee logins, leave assignment and page title checks. It uses assertions, explicit waits, a screenshot utility with a test listener, and a requirements traceability matrix.

Result. The dynamic interface caused frequent NoSuchElement, Timeout and StaleElement errors, which the team fixed by inspecting the page structure, choosing sturdier locators and adding proper waits. The final TestNG run in Eclipse showed all 13 tests passing.

UX Testing: Canadian Tire website

Problem. Where do shoppers struggle when buying from a major retailer's website?

What I did. Team usability evaluation for End User Experience Testing (INFO-6050). Three participants tried tasks such as finding winter tires, filtering products, checking out and locating return policy details. The work included user personas, a heuristic evaluation using the 5Es model and a usability test plan.

Result. Filtering and navigation tested well. Finding return policy details and completing checkout on mobile caused problems.

Data-driven test automation with LLM-generated test data

Two-person project for Automated Test Tools (INFO-6105). Used an open-source LLM through Ollama to generate CSV test data for OrangeHRM's add-employee and employee-contact forms, with strict field rules such as unique five-digit employee IDs, ISO dates and Canadian phone numbers. Built a data-driven Selenium and Java framework with page objects and a CSV reader to run those records through the add-employee form, and set it up to run in Jenkins.

Mobile test automation: Appium on an Android app

In-class project for the Capstone course (INFO-6068). Set up Android Studio, an Android emulator, the Appium server and Appium Inspector, installed the BMI Calculator app and used Inspector to find its element locators (the unit dropdowns, the weight and height fields and the Calculate BMI button). The scenario was to choose a weight unit, enter a weight and height (120 and 165 cm gives a BMI of 44.1), tap Calculate BMI and check that a result appears. I wrote a Java, Maven and TestNG framework with a driver factory and base test class for it, and a reflection on the approach, locator challenges and use of AI tools. The project also holds test classes for the Money Manager mobile app.

Bowling game: BDD and TDD in Java

Two-person project for Testing for Development (INFO-6067) that built a ten-pin bowling scorer using acceptance-test-driven and test-driven development. The Gherkin feature file has five scenarios (a gutter game, all ones, a spare, a strike and a perfect game), run with Cucumber step definitions alongside seven JUnit 5 unit tests, plus a console program that plays all ten frames. Built with Java and Maven.

Load testing: address book web application

Project for Non-Functional Testing (INFO-6055) on a PHP and MySQL address book application. Built an Apache JMeter test plan that recorded the site's pages, including the home page, the list of all entries and the add-entry form, and replayed them under load with response assertions, timers, cookie and cache handling, and test data fed from a CSV file.

Selenium IDE test automation: BestBuy.ca

Team project (TEK Automation) for Automated Test Tools (INFO-6105). Recorded and replayed a Selenium IDE test on BestBuy.ca that searches for a laptop, adds it to the cart and checks that the cart updates. The 13-step test ran with no failures. My part was test case design and execution for product search and navigation.

Research paper: software test patterns

Research paper for Testing for Development (INFO-6067) on what test patterns are, where they fit in unit, integration and system testing, and their benefits and drawbacks. It examines the Test Double pattern (stubs, mocks and fakes) with an example of mocking a payment gateway to test a payment processor without the real service.

Research report: The Beginning and End of Information Systems

Team research presentation on when an organization should adopt an information system and when it should retire one. It uses a what, how and when framework, with examples, effects on business performance and recommendations such as monitoring performance and planning upgrades from the start.

Java: Travel agency class hierarchy

Object-oriented Java project for Coding for Test (INFO-6066). An abstract Travel class with Cruise, RiverCruise, Resort and AllInclusive subclasses, plus a TripDetails class that shows inheritance, method overriding and polymorphism using an array of trip objects.

Java: Lifeboats calculator

Java console program for Coding for Test (INFO-6066). It takes the ship name, number of people, lifeboat capacity and lifeboats available, then works out the minimum lifeboats needed, how many people are rescued or drown (with percentages) and any spare capacity.

Context data flow diagram: Pine Valley Furniture online system

Systems analysis diagram of an online furniture business. It traces a customer order through receiving the order, updating furniture sales and inventory files and producing a sales management report, with the customer, bank, delivery, vendor and admin as external parties.

Test Methodologies: Best Sell sales system (Java, Python and COBOL mainframe)

Problem. A retailer's new Java data entry, Java and Python analytics and COBOL mainframe subsystems had to work with an old third-party data entry system to produce salesperson reports, and the team needed a test approach for each part.

What I did. Wrote an individual test strategy for the system testing phase (black box for the old system, white box for the three new ones, across unit, integration, system, acceptance and regression levels) and a group strategy using a V-Model and risk-based approach with equivalence partitioning, smoke testing, defect reporting and exit criteria. Set up IBM Explorer for z/OS, wrote and ran COBOL compile-link-go JCL, and designed 44 invalid-input test cases for login, salesperson, sales and commission fields.

Result. The COBOL job (JPR1INV1) built the VSAM file, compiled and linked the three COBOL programs, ran the transaction and report programs and printed a transaction log and a six-page report. Every step finished with condition code 0000 except one COBOL compile that returned 0004, a warning, so the job ended with return code 0004. Of the 44 invalid-input cases, 25 passed, 10 failed and 9 were not applicable. The 10 failures were defects in how the system handles invalid input.

Managing Software Engineering Project

Completed in the Managing Software Engineering course of my Quantic Master of Science in Software Engineering.

Software Design and Architecture Project

Completed in the Software Design and Architecture course of my Quantic Master of Science in Software Engineering.

Web Application and Interface Design Presentation

Completed in the Web Application and Interface Design course of my Quantic Master of Science in Software Engineering.

Introduction to Machine Learning Project

Completed in the Introduction to Machine Learning course of my Quantic Master of Science in Software Engineering.

Artificial Intelligence Engineering Presentation

Completed in the Artificial Intelligence Engineering course of my Quantic Master of Science in Software Engineering.

Parking Lot Manager: Python desktop app

Problem. A parking lot with regular and electric vehicle (EV) bays needs a simple way to park, remove and find vehicles, and to see how charged each EV is.

What I built. A Python and Tkinter desktop app on an object-oriented design: a Vehicle base class with Car, Truck, Motorcycle and Bus subclasses, and an ElectricVehicle class with ElectricCar and ElectricBike subclasses that keeps battery charge between 0 and 100 percent. A ParkingLot class creates a floor with separate regular and EV slots and handles parking, removing, full-lot messages and number checks on the inputs.

Result. The app finds a vehicle's slot by registration number or colour, lists registration numbers by colour, shows the current lot status and shows EV charge levels.

Hotel booking data: generator and validator (Java)

Problem. Booking data is only trustworthy if bad records, such as a check-out before check-in or a wrong total price, are caught before they reach reports or billing.

What I did. Used a Java generator that writes 50 hotel bookings (Single, Double and Suite rooms at 100, 180 and 250 per night) to a CSV file, and a Java validator that reads the file and flags any booking where check-in is not before check-out, or where the total price is not the nightly rate times the nights.

Result. Running the validator on the 50 records reported 42 date errors and 8 price errors. A further check of the same file found 48 of the 50 records where the stated number of nights does not match the dates, a case the validator does not test.

PolicyLens-Mini: ask questions of a policy PDF

Problem. Employees need quick answers from long policy documents, and a confident wrong answer to a question the document does not cover is worse than no answer.

What I built. A FastAPI backend that reads an uploaded PDF and answers questions, and a Next.js, React and TypeScript frontend. Answers come from my own retrieval code: policy-section detection, TF-IDF-style weighting, cosine similarity and a relevance threshold that rejects questions the document does not support. The frontend is deployed on Vercel and the backend on Render, with GitHub Actions building both. I worked in three sprints on a GitHub Projects board.

Result. On a 20-question evaluation set the app went from 15 correct to 20 correct after I fixed the retrieval failures the evaluation exposed. The final regression suite of 35 tests passes.

Policy-RAG-App: policy question answering

Problem. Employees need answers about company policies that come straight from the policy text, with the source shown, and not invented.

What I built. My Quantic AI Engineering project (August 2026): a FastAPI backend and a simple web page that answer questions from 16 synthetic company policies. The policies are split by heading, indexed with TF-IDF (scikit-learn) and ranked by cosine similarity. Each answer shows the matching policy section and its source, and questions scoring below 0.25 similarity are rejected so the app does not guess.

Result. Evaluated on 50 questions: groundedness 88%, relevance 86%, correctness 90% and citation accuracy 90% (45 of 50), with median latency of about 720 ms and 95th percentile of about 1.4 seconds. Most errors came from ambiguous questions that touch more than one policy. I also recorded the required demo video.

BinaryShield AI: malware detection web app

Problem. Security teams need a fast first check on whether a Windows executable looks like malware or goodware.

What I built. My Quantic Introduction to Machine Learning project (July 2026). I trained on about 50,000 Windows executable records with 27 numeric features, held out 20% for the final test and chose models with stratified 10-fold cross-validation, fitting the preprocessing only on training folds to prevent leakage. I compared seven model types (logistic regression, decision tree, random forest, a PyTorch neural network, XGBoost, LightGBM and CatBoost) and picked the best by mean AUC and stability. The Flask web app takes an uploaded .exe, reads its header features with pefile, and also scans it with YARA rules. It also supports manual entry and batch CSV prediction.

Result. pytest tests run on every push to main in GitHub Actions, and a deployment to Render only starts when they pass. The app is live with a health-check endpoint.

CI/CD pipelines with GitHub Actions

Problem. Code changes should be built and tested automatically before they reach users.

What I did. Wrote the pipelines in four repositories. A React app builds, runs its tests and deploys to GitHub Pages on every push to main. The malware app tests and then deploys to Render. In a Quantic course app I added automated unit-test and deploy workflows. In a course Docker project I added a workflow that builds the container images.

Result. Each pipeline runs on every push, so a failing test or build is seen straight away.

Ontario College Graduate Certificate

Software and Information Systems Testing

Fanshawe College, London, Ontario. Awarded December 2025.