QA Tester for Secure AI Knowledge MVP

Title: QA Tester for AI Q&A and Document Metadata MVP (RAG / LLM testing) About the project We are building an AI MVP for a UK government-funded knowledge platform on climate, environment and nature. It has two features: a question-and-answer assistant that gives cited answers from a curated document library, external libraries and the web; and an automated metadata extractor for bulk document uploads, reviewed by a communications team before publishing. Security and accuracy are critical.

What you will do

• Write a test strategy and plan • Build a golden set of test questions and expected answers, and score answer accuracy, citations and hallucinations • Test bulk upload and metadata extraction accuracy • Test for prompt injection, data leakage and access-control issues • Run regression, performance, accessibility (WCAG 2.2 AA) and cross-browser checks • Log clear defects and send a short weekly QA report • Produce the MVP acceptance report and support a small user usefulness test Must have • 3+ years of QA on web applications • Proven experience testing LLM or RAG applications • Knowledge of AI security risks (OWASP Top 10 for LLMs) • Strong written English Nice to have • Promptfoo, Ragas, DeepEval or similar; Playwright or Cypress • Accessibility testing • Public-sector or donor-funded project experience Engagement • Remote, part-time (about 20 hours per week) • Mid-November 2026 to end of January 2027, starting with a 2-week trial • At least 2 hours' daily overlap with GMT • NDA required • Budget: To apply, answer these: 1. Describe one LLM or RAG application you tested. How did you measure answer quality? 2. How would you test whether a cited answer is actually supported by its source? 3. Give two prompt injection tests you would run against a document Q&A tool. 4. Which tools would you use for this project, and why? 5. Your availability from mid-November and your rate. Applications without answers to these questions will not be considered.