AI Testing Academy
Prove a non-deterministic system works — and keep proving it.
Evaluation sets, LLM-as-judge, regression gates in CI, red-teaming and drift monitoring. The academy that turns a demo into something you can put a name against.
Duration
5 weeks
Content
38 hours
Lessons
44
Level
Advanced
Students
468
Certified
245
RIVA mentors you through every lesson, lab and assessment in this academy — and knows where you got stuck last time.
What you’ll be able to do
Written as capabilities, not topics. If you can’t do these at the end, the academy didn’t work.
- Build an evaluation set that reflects real traffic
- Use LLM-as-judge without fooling yourself
- Gate a deploy on evaluation regressions in CI
- Red-team a system and monitor for drift afterwards
Skills you'll build
Tools you'll use
5 modules · 44 lessons
Module outline published. Full lesson list unlocks when the next cohort opens — we don't list lessons before they're recorded.
01Eval setsSampling real traffic without leaking anything private.9 lessons · 2h
Lesson list publishes when the cohort opens.
02GradersDeterministic checks, rubrics and LLM-as-judge calibration.9 lessons · 3h
Lesson list publishes when the cohort opens.
03CI gatesFailing a build on a quality regression.8 lessons · 2h
Lesson list publishes when the cohort opens.
04Red-teamingInjection, jailbreaks and abuse cases, systematically.9 lessons · 2h
Lesson list publishes when the cohort opens.
05MonitoringDrift, canaries and production feedback loops.9 lessons · 2h
Lesson list publishes when the cohort opens.
Hands-on labs
Fail a build on a quality regression
Wire an evaluation suite into CI so a prompt change that drops faithfulness never reaches main.
40 min · Advanced · runs in the browser
Projects you can ship
Structured Talent Screening Assistant
Score applications against a written rubric — evidence quoted from the application, protected attributes excluded by construction, and every score reversible.
HR · 22h · Professional
Network Incident Commander
Correlate alarms across a regional network into one incident, propose the likely root cause with evidence, and draft the customer notice — under supervision.
Telecom · 34h · Expert
On completion
ReveloSoft Certified Professional
You design systems other engineers build on, and you can defend the design under pressure. Evaluation, cost and failure behaviour are part of your answer, not an afterthought.
120-minute assessment · pass mark 80% · QR-verifiable credential
- Hold the Associate certificate
- Complete three Professional-level academies
- Pass the case-study and system-design assessment
- Ship an enterprise project with an evaluation suite
- Pass a technical review with a ReveloSoft mentor