APPENDIX H: RESUME PORTFOLIO BUILDING CAPSTONE PROJECTS

HOW TO PRESENT IT IN A PORTFOLIO README outline Title and a purpose sentence mentioning evaluation and observability. The report-card analogy in a few sentences. A diagram in text: dataset to runner to harness to scorer to results to dashboard. How to run everything in order. What each of the five checks means and why it was chosen. What the quality gate does. An embedded screenshot of the dashboard. Honest limits, such as the fact that the models are scripted and the dataset is small. What I would measure next. What to screenshot or record Take a screenshot of the dashboard with all four models. Take a screenshot of the terminal showing the quality gate passing and then failing. Take a screenshot of a few lines of a trace file. Record a short video in which you break the harness on purpose, watch the gate fail, and repair it. Metrics to report honestly Report the pass rate for each model version, which should be ten of ten, three of ten, zero of ten, and zero of ten. Report the number of cases and checks, which are ten and five. Report latency only as a demonstration, noting that the numbers come from a mock model on one computer and tiny local steps, so they say nothing about real latency. State that the dataset is invented and small. How to word a resume bullet truthfully Good example: "Built a Python evaluation suite for an agent harness with a 10-case synthetic dataset, five automated checks, per-call tracing, a pass-rate quality gate, and an HTML dashboard comparing four model versions." Another good example: "Demonstrated a regression gate that fails a build when a small change breaks agent output, using tracing and automated scoring." Avoid saying "benchmarked state-of-the-art models" or any wording that suggests real models were measured, if they were not. Interview talking points Explain the difference between a log, a metric, and a trace. Explain why you built flawed models on purpose. Explain why a pass rate alone can hide problems, and how the per-check breakdown helps. Explain what a quality gate does in a build system. Explain the limits of a small dataset and how you would grow it.