Examiner

Challenge

Examiner began during preparation for Microsoft’s Azure Solutions Architect Expert certification. Practice material gathered from different sources did not always have dependable answers or explanations. That made it hard to tell whether a learner was reinforcing sound knowledge or rehearsing a mistake.

The product problem extended beyond question accuracy. A correct response in one session is useful feedback, but it does not by itself show whether the underlying knowledge will be available later. Examiner treats content trust and learning-state design as connected engineering problems.

Vision

Build a maintainable certification learning system in which questions can be reviewed and corrected as source content, practice can be adapted to a learner’s progress, and future review can reinforce knowledge after initial mastery.

The longer-term direction is a learning lifecycle rather than a single exam attempt. Durable review is a design objective; the current application’s pool-based mastery behavior remains distinct from the proposed scheduled-review model.

Current Solution

Current capability

Technical Architecture

The current architecture separates question authoring and compilation from the learner-facing experience. Azure infrastructure is defined as code; the presence of these definitions should not be read as a claim that every environment is currently deployed or production-operated.

1. Author and verify: certification questions are maintained as structured Markdown source material.
2. Extract and compile: Node.js tooling validates and transforms source content into a compiled question dataset.
3. Render and study: a React and Vite web application loads the dataset and uses SurveyJS to render exam and study interactions.
4. Persist study state: .NET isolated Azure Functions expose APIs for adaptive stats, study sessions, tags, and adaptive pools; Azure Table Storage is the configured persistence service.
5. Define the platform: Bicep describes Azure Static Web Apps, Functions, Storage, Key Vault, and Application Insights with Log Analytics.

Current data model and APIs support the existing progress and pool behavior. The separate v2 due-review tables, endpoints, and migration path are design work, not represented as live services here.

Learning Model

Key Design Decisions

• Keep authored content outside the UI: source material can be reviewed and transformed without coupling every correction to application code.
• Preserve question structure: different interaction types need explicit representations through extraction, compilation, and rendering.
• Separate session progress from knowledge state: finishing a bounded session and advancing a question’s mastery are different events.
• Make pool behavior explicit: pool membership is a short-term selection mechanism, not a synonym for durable mastery.
• Design migrations before changing the learning model: the proposed v2 path accounts for compatibility, backfill, validation, and rollback rather than assuming existing learner state can be discarded.

Technologies

Lessons Learned

• Content quality is part of product reliability: an incorrect answer can undermine an otherwise well-designed study flow.
• Structured content needs contract-preserving transformations; parsing, normalization, compilation, and rendering all affect the learner’s result.
• Session completion is easy to count, but it is not evidence of durable learning. Product metrics should reflect the claim being made.
• A scheduled-review feature changes storage access patterns as well as UI. Due dates and event history need to be modelled and queried intentionally.
• Roadmap and implementation must be labelled separately. In this project, the v2 spaced-review model is explicitly a design, not a shipped feature.