Introduction
I set out to build a small teaching app that helps students feel how mortgage stress testing changes what they can borrow. The widget invites them to enter a home price, down payment, amortization, and carrying costs, then compare their contract-rate results to a stressed qualifying rate. It calculates GDS (housing costs divided by income) and TDS (housing plus other debt divided by income), shows Pass / Caution / Fail badges, and—importantly—reveals which ratio is binding. The goal is less “calculator,” more conceptual x-ray: students see how a small rate bump can push ratios past thresholds, and why the same borrower can pass or fail depending on taxes, heating, condo fees, or other debt. CMHC’s definitions include 50% of condo fees in the numerator for GDS/TDS, which the tooltips explain in plain language (CMHC, 2018, 2020).
I used my mobile phone in the development of this app, using AI chats along with using Firebase Studio on my mobile phone. I utilized a desktop computer as well, given easier development with a larger screen!
You can access the calculator here. Please note there may be errors in the app, as this was a pretty rapid protoype.
Figure 1. Rate Stress Test & Affordability Calculator (full UI).
Because students often ask whether the stress test is “real,” I made the policy explicit and configurable: by default the qualifying rate is the greater of the contract rate plus 2% or 5.25%, mirroring federal guidance for many uninsured mortgages, with the caveat that policy can vary by product and over time. I also note the 2024 OSFI change that removed a set minimum qualifying rate for uninsured straight switches at renewal (no higher balance or longer amortization). In other words, the app is a learning model, not lender‑exact underwriting (OSFI, 2024a, 2024b).
How I Built It (Firebase Studio + Gemini)
Firebase Studio is Google’s browser-based, AI-assisted builder inside the Firebase console that turns plain-English briefs into running apps. Powered by Gemini’s App Prototyping agent, it proposes architecture, scaffolds real React/TypeScript code, and wires common Firebase services—Hosting/App Hosting, Authentication, Firestore, and Storage—while letting you iteratively refine the project in natural language. When you need generative features, you can connect the Gemini API from the same workspace, then preview and publish directly to Firebase Hosting without context switching or lock-in. For educators, this shortens the path from instructional idea to classroom artifact: prototype a learning tool, embed it, and later add analytics or auth in the same ecosystem (Google Cloud, 2025; Google, n.d.-a, n.d.-b, n.d.-c).
Why the Design Looks the Way It Does
I focused on a few research-backed principles rather than piling on features. First, timely, actionable feedback matters. The pass/caution/fail badges and the “binding ratio” label operationalize Hattie and Timperley’s three feedback questions—Where am I going? How am I going? Where to next?—right in the interface (Hattie & Timperley, 2007). When a student nudges the rate slider, the answer updates immediately with the reason (e.g., “GDS now exceeds 39%”), which nudges metacognition rather than mere number‑watching.
Second, I tried to reduce extraneous cognitive load while respecting the inherent complexity of mortgages. Tooltips keep definitions beside the controls; the layout chunks the task into “Inputs → Contract results → Stressed results → Maximum affordability.” That blend follows Cognitive Load Theory—manage working memory by removing avoidable friction—and Multimedia Learning guidelines such as segmenting and coherence (Mayer, 2020; Sweller, 2019). The result is a page you can scroll on a phone without context‑switching to a glossary or another tab.
Third, I designed the interaction to promote constructive engagement. Students don’t just read about GDS/TDS—they manipulate inputs, compare contract versus stress, and explain what changed. That nudges them along the ICAP spectrum (Passive → Active → Constructive → Interactive), which predicts deeper learning as learners generate and test their own ideas (Chi & Wylie, 2014).
Figure 2. Info icon tooltips provide concise definitions at point of need.
Designing Small Apps With AI in Mind
Even though this model is deterministic, I still borrowed from human‑AI interaction because many teaching tools now mix rules and generative output. Two guidelines I value from Amershi et al. (2019) are to make capabilities and limits explicit and to support efficient dismissal or override. In practice that means: (a) clear language that this is a teaching approximation with adjustable policy levers, and (b) controls to change the compounding mode, ratios, and thresholds rather than hiding those details. If I add an “Explain my scenario” button powered by Gemini, I’ll follow the People + AI Guidebook patterns—state uncertainty, cite the math the explanation used, and show concrete next steps (PAIR, n.d.). At the program level, I plan to document any generative feature with the NIST AI Risk Management Framework—intended use, inputs/outputs, known failure modes, and evaluation plan—so instructors can adopt or decline features with eyes open. This also aligns with Shneiderman’s human‑centered AI posture: keep human control high, design for auditability, and treat explanations as part of the product (NIST, 2023; Shneiderman, 2022; Google PAIR, n.d.). Finally, if I extend the app to generate formative feedback text, I’ll map those prompts and outputs to AI literacy competencies—interpreting outputs, recognizing limitations, and reflecting on decisions (Long & Magerko, 2020).
Policy Context and Caveats Students Should Know
Two persistent confusions surfaced during testing. First, policy drift: headlines about the “stress test changing” often mix multiple topics. The minimum qualifying rate for many uninsured mortgages is still framed as the greater of contract plus 2% or 5.25%, yet OSFI carved out a 2024 exception for uninsured straight switches at renewal (no higher balance or longer amortization). The app keeps this honest by labeling the stress‑rate policy clearly and letting instructors set it per lesson (OSFI, 2024a, 2024b). Second, what counts in GDS/TDS numerators is not always intuitive. Students may miss that 50% of condo fees belongs in both GDS and TDS; the tooltip spells it out and links to the definition. The thresholds many students hear—39% GDS / 44% TDS—are common teaching anchors for insured lending, but real underwriting can vary. The app shows the math and the levers so learners can reason from first principles rather than memorize one set of numbers (CMHC, 2018, 2020).
Summary
This post is designed for mobile reading and published to the course blog under (A1) Analyses. It presents an original, media‑based analysis of a small product in the mobile intelligence ecosystem (the widget is embeddable, responsive, and works on phones), and it evaluates its educational merit with explicit links to learning science and human‑AI design research. It also reflects my professional context building mortgage‑related learning resources.
References
Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., … Horvitz, E. (2019). Guidelines for human–AI interaction. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Paper 3). https://doi.org/10.1145/3290605.3300233
Canadian Mortgage and Housing Corporation. (2018, March 31). Calculating GDS/TDS. https://www.cmhc-schl.gc.ca/professionals/project-funding-and-mortgage-financing/mortgage-loan-insurance/calculating-gds-tds
Canadian Mortgage and Housing Corporation. (2020, June 5). What are the general requirements to qualify for homeowner mortgage loan insurance? https://www.cmhc-schl.gc.ca/consumers/home-buying/mortgage-loan-insurance-for-consumers/what-are-the-general-requirements-to-qualify-for-homeowner-mortgage-loan-insurance
Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823
Google. (n.d.-a). Get started with the App Prototyping agent | Firebase Studio. Retrieved October 12, 2025, from https://firebase.google.com/docs/studio/get-started-ai
Google. (n.d.-b). Build an app with the Gemini API | Firebase Studio. Retrieved October 12, 2025, from https://firebase.google.com/docs/studio/build-gemini-api-app
Google. (n.d.-c). Gemini in Firebase. Retrieved October 12, 2025, from https://firebase.google.com/docs/ai-assistance/gemini-in-firebase
Google Cloud. (2025, April 9). Firebase Studio lets you build full‑stack AI apps with Gemini. https://cloud.google.com/blog/products/application-development/firebase-studio-lets-you-build-full-stack-ai-apps-with-gemini
Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81–112. https://doi.org/10.3102/003465430298487
Long, D., & Magerko, B. (2020). What is AI literacy? Competencies and design considerations. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, 1–16. https://doi.org/10.1145/3313831.3376727
Mayer, R. E. (2020). Multimedia learning (3rd ed.). Cambridge University Press. https://www.cambridge.org/highereducation/books/multimedia-learning/FB7E79A165D24D47CEACEB4D2C426ECD
National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0) (NIST AI 100‑1). https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10
Office of the Superintendent of Financial Institutions. (2024a, November 21). Minimum qualifying rate for uninsured mortgages. https://www.osfi-bsif.gc.ca/en/supervision/financial-institutions/banks/minimum-qualifying-rate-uninsured-mortgages
Office of the Superintendent of Financial Institutions. (2024b, November 21). Backgrounder: Minimum qualifying rate (MQR). https://www.osfi-bsif.gc.ca/en/news/backgrounder-minimum-qualifying-rate-mqr
Shneiderman, B. (2022). Human‑centered AI. Oxford University Press. https://global.oup.com/academic/product/human-centered-ai-9780192845290
Sweller, J. (2019). Cognitive architecture and instructional design: 20 years later. Educational Psychology Review, 31, 261–292. https://doi.org/10.1007/s10648-019-09465-5
People + AI Research (PAIR). (n.d.). People + AI Guidebook. Retrieved October 12, 2025, from https://pair.withgoogle.com/guidebook/
