✓ Tier 2 · the flagship
QA Foundation Sprint
You’re shipping AI features on prompt-and-pray. There’s no eval suite, no release gate, and no one who owns whether an output is safe to ship.
In 30–45 days I install your AI-quality system: evals and prompt regression wired into CI, injection-safety and cost guardrails, and a QA operating model your team will actually follow. Playwright and CI come included as the foundation, the eval and AI-testing layer is the value.
Book a call✓ what you get
An installed AI-quality system, and the IP to keep it.
- AI testing layer (the value): golden datasets, prompt regression suites, AI output validation, graceful-degradation checks, prompt-injection safety tests, and API cost monitoring, all wired into CI as release gates.
- Eval design and error analysis: what “good” means for your outputs, encoded as scored evals you own. This is the judgment layer tools can’t give you.
- Automation framework: Playwright (and pytest where it fits), structured for your stack. Included foundation, not the headline.
- CI integration: GitHub Actions or your CI, with smoke + regression gates so a bad change can’t merge quietly.
- Critical-path coverage: automated tests for your top user journeys and API layer, scoped to risk.
- QA Operating Model: roles, rituals, bug triage, release sign-off, and quality gates your engineers will actually follow.
- Reporting: dashboards so quality is visible to you and your board, not buried in a CI log.
- Team handoff: documentation plus two working sessions, so the system outlives the engagement.
✓ who it's for
For teams shipping AI features without a safety net.
- You have AI features in production (chat, agents, RAG, copilots) and no systematic way to test them.
- You’re shipping fast and the lack of a safety net is starting to cost you real time.
- You want an installed quality function in weeks, not a six-month QA hire ramp.
- You’d rather own a reusable eval/test IP set than rent hours of manual testing.
Explicitly out of scope
- Open-ended feature testing after handoff (that’s the Fractional retainer).
- Rewriting product code, I test it, I don’t refactor it.
- A guarantee of zero bugs. Testing reduces risk; it never eliminates it.
✓ the outcome
From prompt-and-pray to a release you can trust. You ship with evals as gates, a framework your team runs, and senior judgment encoded into how you decide what’s safe to ship.
✓ questions
Before you book.
Isn’t a Playwright suite something the funded tools already do?
The framework and CI plumbing are increasingly commodity, that’s exactly why they’re the included foundation here, not the headline. What doesn’t commoditize is deciding what to test, designing the evals, doing error analysis, and owning the “is this safe to ship” call. That judgment layer and the reusable IP you keep are the actual deliverable.
How is this priced?
Fixed at $15,000 for a typical scope, with a range of $12k–$20k depending on how many surfaces and AI features are in play. I price the scope, not the hours, you know the number before we start.
What if we’re too early for a $15k sprint?
Then start with AI QA in a Box, the same system, self-served. It’s the rubrics, scaffolds, checklists, and playbook I install here, packaged as a $300 pack you run yourself. When you’re ready to have it installed and tuned to your stack, that’s this sprint.
What if our scope is bigger than a standard sprint?
We scope it honestly up front. If it’s genuinely larger (multiple AI features, heavy compliance), the price moves within the range and the timeline is set realistically, I’d rather sell a true 6-week timeline than a 30-day one I can’t deliver.
Do we need the audit first?
Not required, but it’s the cleanest entry. The audit produces the roadmap that scopes the sprint precisely, so there are no surprises. Many teams do the audit, then convert it into the sprint.
What happens after the 30–45 days?
You own everything: framework, evals, operating model, and docs. Most teams keep me on as a Fractional QA Lead to maintain and extend it, but that’s a separate choice, not a lock-in.
One bad regression away from a lost week.
Book a 20-minute intro call. I’ll tell you honestly whether I can help and what the right next step is: audit, sprint, or nothing yet.