Work / Case study
License Exam Prep
Designed and shipped an offline Python desktop application that prepares insurance agents for Florida state licensing exams across six license tracks.
Source code is private. Walkthroughs are available on request.
Business impact
A self-serve exam-prep product built to shorten the path from new hire to licensed, revenue-producing agent. It runs on locked-down, carrier-managed workstations with no installs, add-ins or cloud services, and gives each learner a projected score and per-domain analytics.
Problem
Licensing is a revenue gate: new agents can't sell until they pass the state exam. The agency had no practice tool that could run on locked-down, carrier-managed workstations with no add-ins, macros or cloud installs.
Solution
Owned the product end to end, from requirements to deployment. The result is a single-file Windows executable (Python, PyWebView, SQLite, ReportLab). It runs full-length timed exams mapped to the official Pearson VUE Florida content outlines, and adds a projected-score gauge, per-domain mastery analytics and branded PDF performance reports.
Results & highlights
- Built a 1,467-question, outline-mapped item bank (500 hand-written master items) across six license tracks: 2-20, 20-44, 4-40, 2-15, 2-14 and 2-40
- Architected blueprints as data: exam weights are validated at startup, so outline changes need no code change or redeploy
- Integrated an LLM (Anthropic API) for AI-generated study narratives, with the API key DPAPI-encrypted at rest and a deterministic offline fallback
- Quality gate: 17 automated tests passing (verified 2026-09-22)
Competencies demonstrated
- Product ownership: Owned requirements through deployment of a single-file Windows desktop application.
- Software engineering (Python): PyWebView interface, SQLite attempt history, ReportLab PDF reporting and PyInstaller packaging.
- Data-driven design: Exam blueprints stored as data and validated at startup; blueprint-weighted sampling across six license tracks and a 1,467-question bank.
- Learning analytics: Projected-score gauge and recency-weighted mastery by content domain.
- AI integration & security: Optional Anthropic API study narratives, with the API key DPAPI-encrypted at rest and a deterministic offline fallback.
- Quality assurance: 17 automated tests passing.
Architecture
- Blueprints as dataexam weights validated at startup
- Weighted assemblersamples the item bank
- Desktop UIoffline
- Attempt historySQLite
- Mastery projectionrecency-weighted
- ReportsPDF + optional AI narrative
Notes
- Question content is original training material written to the public content outlines; it is not sourced from secured exam items.
- Logo files in
web/assets/are neutral placeholders.