The bank has said it openly
DBS is phasing out about 4,000 contract and temporary roles over three years as AI takes on work previously done by humans.
A hands-on masterclass for QA and test engineers facing non-deterministic AI output.
Pass and fail stop being binary when the system writes answers instead of returning values. Three days on LLM-as-judge, benchmarks, and regression for output you cannot pin down.
“He holds the sole-inventor patent for LLM agent systems, granted in 2023, and has spent close to two decades delivering technical training. He has hand-built AI systems for classified environments where the public internet does not exist. In this masterclass, he shares the architecture, his working methods, and the deployment blueprints drawn from completed projects, compressed into 3 days of intensive, hands-on learning.”Prebook the QA Masterclass
Cohorts of 10 to 25, and seats are running out. Prebooking is free; you pay by invoice only when your cohort is confirmed.
The Real Problem
Classic test cases compare expected and actual. Generative output makes both sides move.
You can cover every branch and still ship nonsense answers, because the risk lives in the language.
A prompt tweak can break behaviour nobody thought to test, and nothing in the diff looks dangerous.
Engineers blame the model, product blames the prompt, and users just leave.
Hundreds of courses, the same recycled content. The trainer sat through them too, and now teaches exactly the part they all skip.
Certificates confirm attendance. The concepts that matter live only in the trainer's head. Here, they live in yours.
Do these sound familiar? Evals are the fix that makes quality testable.
The Trainer
The playbook behind every masterclass was bought with years of production attempts and hard failures. The count of the wall is the lesson, and it is the first thing he teaches.
More than a hundred students in Singapore completed this similar AI training this year. The format is proven with a live audience.
The sole-inventor patent for LLM agent systems, granted in 2023. The legal and logical foundation of modern agents.
18 years of compute on GPUs, back when the field was called HPC, long before AI became a marketing buzzword.
Teaching engineers has been the day job for more than twenty years. Training is the craft, not a side engagement.
25 years across AWS, Azure, GCP, and on-premises bare metal, including air-gapped environments few engineers ever see.
Small cohorts keep the help personal. Seats are running out fast.
Past students say the most valuable part is simple. You get to ask a human who has done a few things successfully before.
Why Now
DBS is phasing out about 4,000 contract and temporary roles over three years as AI takes on work previously done by humans.
About 4 in 10 of those affected by the GovTech cuts were officers in their 40s.
75 per cent of all new code at Google is now AI-generated, reviewed and approved by engineers.
40 per cent of the workforce will need reskilling within three years because of AI.
AI now helps read CT scans, yet imaging demand grew even faster and radiologist vacancies stayed high. When AI makes something cheaper, the world orders more of it.
Klarna let an AI assistant do the work of 700 service agents, then rehired people when quality slipped. The winners are the humans who can direct the AI.
None of these companies failed. They restructured, and loyalty no longer shields anyone. PwC’s 2026 jobs barometer finds roles professionalised by AI growing twice as fast as the rest. Be the one directing the machines, not the one replaced by them.
The Curriculum
| Day | Morning | Afternoon | Focus |
|---|---|---|---|
| Day 1 | You get hands-on with model behaviour and where generative systems surprise teams. | The afternoon covers LLM-as-judge, and how to design a judge you would trust. | How AI Systems Behave |
| Day 2 | You build benchmark sets that represent real users, not cherry-picked cases. | Regression for non-deterministic output follows, so a prompt tweak cannot quietly break behaviour. | Benchmarks and Regression |
| Day 3 | You add AI security testing basics, including prompt injection cases in your suite. | The capstone is an eval suite for a real application, built by you and wired into the delivery loop. | Security and the Gate |
Each day comprises approximately 25% guided sessions, 60% hands-on labs on a real task, and 15% structured reflection. The table above is an example. Actual content includes more practical material than described, may differ slightly as it tracks the latest industry trends, and is refreshed right up to each cohort. The exact setup guide will be shared one week before Day 1.
Before You Arrive
Standard or Custom
Standard
Custom
Prebooking is free. You pay by invoice only when your cohort is confirmed.
The Outcomes
An LLM-as-judge you designed and can defend.
Benchmark sets that represent real users.
Regression testing that works when output moves.
Injection and abuse cases inside your suite.
Criteria that make output quality somebody's job again, and that somebody is you.
A suite wired into the delivery loop for a real app.
Walk out owning the quality nobody else could measure.
Masterclass Menu
The Foundation Course is the shared entry, running every weekend. Each masterclass below runs as a 3-day weekday specialist course, in a Standard open cohort or a Custom private cohort for your organisation.
Technical Track
Technology Track
Manager Track
Leader Track
Functional Track
OPC Track
You bring the domain expertise. We bring the engineering discipline. If you cannot test it, you cannot ship it.
Prebook
Prebooking is free and takes two minutes. You get 48-hour priority access to the next cohort of 10 to 25 each, and we confirm dates, venue, and your invoice before any payment. Built for QA and test engineers whose test plans just met generative output.
This is not only about confidence with AI. It shapes your job, your career, your team, your projects, and the example your children grow up with. Choose the building side while the choice is still yours.
Prebook My Seat Now