Overview
Over roughly 28 hours (2026-07-08 15:06 to 2026-07-09 19:31), this portfolio went from an empty repository to a deployed, evidence-gated platform: a FastAPI backend, a static frontend, an AI chat assistant constrained to public evidence, and a recruiter-facing proof surface -- then through multiple rounds of QA hardening driven by real regression testing rather than assumption. The citation source for this entire case study is the commit ledger covering that window (76 commits, first commit 7a9590a4).
Starting Point
The first three commits on 2026-07-08 at 15:06 established the whole platform's skeleton in minutes: 7a9590a4 ("add FastAPI backend + Vercel/GitLab deployment config", 58 files, 5,422 insertions), a8b0771a ("add static frontend site", 27 files, 4,735 insertions), and b8b2d6a4 ("add canonical résumé and archive-only variants"). This is a from-scratch build, not an iteration on existing code.
Buildout Phase
The following ~24 hours added features in discrete, single-purpose commits: a four-slot AI News Ticker (ce1f8a7f), public polish and a chat canary gate (aabfdb22), an EN/ES bilingual toggle (b4eae0ef), a Resume Match Explainer (2aa2ef5e), a JD Fit Score feature (2560da8b), the Career Evidence Platform Trust Gate (0bf047ed), live provider health checks (54d5f602, 5fa96f0a), an Enterprise Deliverables case-study page (d1707ce7), an autoplaying hero video ported from an existing RetailVitamin pattern (f661ee5a), and a Go evidence-normalizer proof slice (359e7732).
Hardening Phase
The back half of 2026-07-09 shifted from adding features to closing gaps found by direct QA: renaming and re-gating proof labels across app/frontend/tests (49b985fc, 60da5167, 41fc79d6), repeated mobile hero-media repair (45ceedaa, c1314297, b81c48f2), capture-slice stabilization (622253a4, eb129f19), SEO/browser-preview polish (e68282b0, b5780420), a perception-layer QA pass (99386c1a), and wiring real Microsoft certification proof (c2ddd1a5).
The chat assistant itself went through the same discipline: fixing over-rejection and bounding skill routing (bce78271, 17666e70), answering public AI-tooling and years-of-experience questions instead of rejecting them (36ca5d16, d9e9c27d), adding a dedicated Guardrail Validation surface (93e2f166), and building structured employment-sequence answering (3b52941c).
Verification Phase
The final two commits in this ledger (2f6f8246, 45ea0633) closed gaps found specifically by running a live regression sweep against the deployed chat assistant on production -- not just unit tests against local code, but real HTTP calls against nkosifelix.com/chat checked against the evidence ledger for correctness. That live-sweep discipline continued past the end of this ledger file as an ongoing practice, not a one-time event.
What This Proves
- Ship velocity with discipline, not chaos. 76 commits in ~28 hours, each scoped to one concern, with commit messages that describe what changed and why.
- Regression-tested public claims. The hardening and verification phases are driven by real QA findings (mobile capture regressions, live chat-sweep failures), not guesswork.
- Operator-grade delivery on a public system. The same evidence-gating discipline used for a portfolio chat assistant elsewhere in this repo (see Hiring Manager Proof) is visible here in the commit history itself.
Evidence Needed / Open Proof Gaps
Per this site's own evidence-first standard, these gaps are named explicitly rather than glossed over:
- Verification needed: add screenshot.
public/proof/screenshots/home-mobile.png,public/proof/screenshots/chat-guardrail.png. - Verification needed: add test output.
public/proof/test-output/latest-regression.txt. - Verification needed: add health-check capture.
public/proof/health/latest-provider-health.json. - This case study covers only the commits present in
commit-ledger-from-7a9590a4.txt(through commit45ea0633, 2026-07-09 19:31). Work after that point is not reflected here.