About the role

Data Wow builds AI and data systems for clients in finance, energy, and retail. Much of that work now involves language models: document extraction pipelines, retrieval-based assistants, classification and summarisation services, and agents that call internal tools. These systems fail differently from the software our industry learned to test. The same input can produce a different answer twice, correctness is often a matter of degree, and a change to a prompt or a model version can quietly move accuracy across the whole product.

This role exists to make that testable. You will build the automated test and evaluation layer that sits underneath our client delivery work, so that a team can change a prompt, swap a model, or reindex a knowledge base and know within minutes whether quality moved. Alongside that you will cover the conventional ground these systems still run on: APIs, data pipelines, web front ends, and the CI that ties them together.

You will work inside project teams rather than beside them. Most of your time is spent on client engagements with engineers, data scientists, and delivery leads, and part of it is spent turning what worked on one project into something the next team can pick up without starting over. This is the first hire dedicated to test automation for our AI delivery work, so you will have real influence over how we approach it.

Key responsibilities

• Build evaluation suites for LLM and ML features: curated test sets, scoring against reference answers, LLM-as-judge where it is justified, and human review where it is not.

• Define what "good enough" means with the delivery team before launch, in numbers a client can accept, and hold releases to it.

• Test retrieval pipelines end to end, covering chunking and indexing, retrieval quality, grounding of the final answer, and behaviour when the right document is simply not there.

• Write regression tests that tolerate non-deterministic output, using semantic similarity, structured-output validation, tolerance bands, and repeated sampling rather than exact string matching.

• Cover failure modes that matter to clients: hallucination, prompt injection through user or document content, leakage of data across tenants or sessions, refusal and over-refusal, and degradation on Thai-language input.

• Track cost and latency alongside accuracy, so that a quality improvement that triples token spend is visible before it ships.

• Detect drift after release by comparing production samples against the baseline, and raise it early rather than after the client does.

• Design and maintain automated test suites for APIs, services, data pipelines, and web applications built for client projects.

• Run tests in CI, keep them fast and stable, and treat a flaky suite as a defect in its own right.

• Set the test strategy at the start of an engagement: what gets automated, what stays manual, what the exit criteria are, and what evidence the client will receive.

About you

• 4-5 years in test automation, QA engineering, or software engineering with substantial testing ownership.

• Strong programming ability in Python, or in TypeScript or Java with the willingness to work primarily in Python.

• Hands-on experience testing an LLM or ML-based product in production or close to it.

• Practical command of at least one automation framework such as Pytest, Playwright, Cypress, or RestAssured, and experience running suites in CI.

• Working knowledge of API and data testing, including SQL and validating data through a pipeline.

• The judgement to decide what deserves automation and what does not, and to say so when a suite is producing noise rather than signal.

• Clear written and spoken communication in Thai and English.