ML Engineer - Retrieval & Grounded Generation

Key Responsibilities

Test Strategy and Acceptance Criteria

• Own the system test plan covering functional, integration, performance, and load testing, and maintain it as a formal deliverable
• Translate the acceptance criteria set by product into executable tests
• Review acceptance criteria with product before development begins, and identify requirements that are ambiguous or not verifiable as written
• Hold engineering, platform, and partner teams accountable to the acceptance criteria, and define what evidence must exist before a change is considered complete
• Work alongside the engineering team day to day, close enough to test their work well and independent enough to report what the testing shows
• Plan and coordinate user acceptance testing with the customer, and track findings through to resolution

Test Automation and Coverage

• Build and maintain automated test suites, and integrate them into the CI/CD pipeline as release gates
• Establish and hold an automated coverage standard, and report against it
• Build performance and load testing that validates the system under its expected user load
• Maintain regression coverage across releases

Measurement and Evidence

• Build and maintain labeled evaluation sets, working with subject-matter experts whose time is limited
• Design and run sampled audits confirming that automated decisions were correct
• Measure and report what the system misses alongside what it gets right
• Measure how often users override the system, and report the trend
• Document how each committed measure is computed, on what data, and what evidence accompanies a release
• Produce this evidence independently of the people building the models

Release Governance

• Maintain the version inventory across builds, models, configurations, and thresholds
• Produce test reports, evaluation records, and rollback criteria for each release
• Work with platform engineering to confirm that test environments match production closely enough for results to be meaningful
• Hold the gate: no release ships without the evidence its acceptance requires

Required Qualifications

• 5+ years in software quality engineering, test automation, or software engineering with substantial dedicated test ownership
• A clear point of view on how testing should be done, and the judgment to know which practices are worth holding to. You will be setting the standard here rather than following one
• Strong coding ability in a language and framework used for test automation, for example Python with pytest, or JavaScript or TypeScript with Playwright, Cypress, or Selenium. Python is what most of this stack is written in. You should also be able to read and reason about the application code under test
• Routine use of AI-assisted development, with informed judgment about where it adds value and where its output requires verification. We expect AI tooling to be part of how test coverage gets built here, not an occasional experiment
• Demonstrated experience building automated test suites and integrating them into CI/CD as release gates
• Experience owning a test plan through review and acceptance
• Demonstrated habit of testing against the requirement rather than the implementation, including working directly with product owners or end users to establish what correct actually means
• Experience with performance and load testing, including establishing what a result does and does not demonstrate
• Working statistical literacy, including sampling design and the confidence a measurement can support
• Experience measuring what a system misses, not only what it gets right
• Willingness to report a result the team did not want, and to keep the measurement independent of the people whose work it evaluates
• Ability to gate work owned by people who do not report to you, and to earn enough credibility with engineers that the gate holds
• Clear, audit-ready written documentation, since much of the output of this role is evidence that others read
• Willingness to hold a release, and explain the reason, to people who want it shipped
• US Citizenship Required
• Active US Secret clearance. The work is performed in a controlled government cloud environment and requires a favorable investigation and CAC eligibility from the start
• Elevated security requirements apply to portions of this work and are discussed during the interview process
• Willingness to travel up to 25% to customer sites, DEFCON AI HQ, and vendor facilities as required

Preferred Qualifications

• Clearance: active Top Secret
• Environment: government DevSecOps, RMF, or ATO environments, and experience producing test evidence for a formal accreditation or audit
• Automation: contract and API testing, test data management, and building test infrastructure others can extend
• AI-assisted testing: using language models to generate and maintain test coverage, and the practices that keep generated tests meaningful rather than merely passing
• Performance: JMeter, k6, Locust, or comparable
• Evaluation: validating machine learning or model output, including groundedness and citation checking, and analyzing where users disagreed with a system recommendation
• Domain: systems where an incorrect result carries real cost, and where testing had to satisfy an outside reviewer rather than only an internal one
• Governance: model cards, NIST AI RMF, or comparable responsible AI practice
• Accessibility: Section 508 and WCAG verification

What Success Looks Like

• A test plan the customer accepts, and that holds up as the system grows
• Acceptance criteria that are verifiable before development starts
• Automated coverage meeting its standard and running as a release gate
• Sampled audits designed to catch an incorrect automated decision
• Performance evidence that reflects realistic load
• A release trail complete enough to reconstruct why any version shipped