Full Description

Armeta is an applied-AI company building engineering-intelligence products for construction and oil & gas. We're hiring our first QA.

Today we have one product in production for a government client, two more approaching release, and four modules ahead of us over the next six months. There is no QA function: testing happens ad hoc, done by developers and analysts, driven by whatever surfaced last. You're not joining a process. You're building one, and then hiring the team around it.

The cost of a defect here is higher than usual. We go through formal acceptance testing and information-security certification. A bug found by the customer during a demo doesn't cost a hotfix — it costs a formal disagreement protocol.

WHAT YOU'LL BUILD

Own product quality end to end: test strategy, process, standards, documentation, metrics. From zero.

Build requirements traceability: spec clause → test case → run result → evidence for acceptance. The acceptance package should assemble itself from the system, not get written by hand two weeks before delivery.

• Author the test program and methodology, run acceptance alongside the analysts and the customer, own the protocols.

• Test an ML system as an ML system, not as a regular backend. Classifier, OCR, and extraction quality is measured with metrics on golden datasets. You own those datasets, set the thresholds, and make sure a new model version doesn't regress on old cases.

• Build automated tests in CI: API, integration, regression. A pyramid, not a hundred flaky e2e scenarios.

• Test business scenarios, not just "the code ran". Most of our defects are gaps between what the spec says and what the developer understood.

• Load testing against SLA: heavy document processing, pipeline throughput, behaviour under queue pressure.

• Test in a closed network perimeter and on-prem, including environments with no internet access.

• Cover security requirements as part of certification prep: requirements checklist, access control, audit logging, secret handling.

• Establish defect culture: a single tracker, severity criteria, DoD that includes tests, and post-mortems on anything that reached production.

REQUIREMENTS

• 3+ years in QA, including building a process or leading a function. A large team isn't required — what matters is that you've built, not only executed.

• Python and pytest at a confident automation level: API tests, fixtures, parametrization, mocking.

• Test design fundamentals: equivalence classes, boundary values, test cases readable by someone other than their author.

• REST and WebSocket APIs, background jobs and queues — you understand how to test asynchronous processing and what idempotency means.

• PostgreSQL and SQL good enough to verify data yourself.

• CI/CD: running tests in the pipeline, merge gates. Docker, Kubernetes required.

• Experience testing ML systems or anything with non-deterministic output, or the ability to get there fast: you understand the difference between pass/fail and a quality metric.

• Experience with formal acceptance: acceptance testing, test documentation, working with an external customer.

• Ability to hold a technical conversation with a non-technical engineer, take criticism without getting defensive, and stay composed when something breaks during a demo.

NICE TO HAVE

• Government customers and GOST documentation standards.

• Participation in information-security certification.

• Testing OCR, computer vision, or document data-extraction systems.

• Load testing: k6, Locust, JMeter.

• Basic security testing: OWASP, access control verification.

• Deploying and testing in restricted-network or on-prem environments.

• Engineering background, or domain experience in construction, design, or oil & gas.

HOW WE'LL MEASURE YOU Not by the number of bugs filed. By whether acceptance passes on the first attempt. By whether regression catches a defect before the customer does. By how long it takes from a problem spotted in the field to a closed test case in the regression suite. And by whether, six months from now, the delivery documentation package takes a day to assemble instead of two weeks.