Functional AI Tester - GenAI

- - - - - - - - - - - -

About the Role

You will be involved in QA for GenAI features including Retrieval-Augmented Generation (RAG), conversational AI and Agentic evaluations. The role centers on:

• Systematic GenAI evaluation (qualitative and quantitative metrics)

• ETL and data quality testing for the data flows that feed AI systems

• Python-driven automated testing

This position is hands-on and collaborative, partnering with AI engineers, data engineers, and product teams to define measurable acceptance criteria and ship high-quality AI features.

Key Responsibilities

• Test strategy and planning

• • Define risk-based test strategies and detailed test plans for GenAI features.

  • Establish clear acceptance criteria with stakeholders for functional, safety, and data quality aspects.

• Python test automation

• • Build and maintain automated test suites using Python (e.g., PyTest, requests).

  • Implement reusable utilities for prompt/response validation, dataset management, and result scoring.

  • Create regression baselines and golden test sets to detect quality drift.

• GenAI evaluation

• • Develop evaluation harnesses covering factuality, coherence, helpfulness, safety, bias, and toxicity etc.

  • Design prompt suites, scenario-based tests, and golden datasets for reproducible measurements.

  • Implement guardrail tests including prompt-injection resilience, unsafe content detection, and PII redaction checks.

  • Track quality metrics over time.

• RAG and semantic retrieval testing

• • Verify alignment between retrieved sources and generated answers.

  • Verify adversarial tests.

  • Measure retrieval relevance, precision/recall, grounding quality, and hallucination reduction.

• API and application testing

• • Test REST endpoints supporting GenAI features (request/response contracts, error handling, timeouts).

• ETL and data quality validation

• • Test ingestion and transformation logic; validate schema, constraints, and field-level rules.

  • Implement data profiling, reconciliation between sources and targets, and lineage checks.

  • Verify data privacy controls, masking, and retention policies across pipelines.

• Non-functional testing

• • Performance and load testing focused on latency, throughput, concurrency, and rate limits for LLM calls.

  • Cost-aware testing (token usage, caching effectiveness) and timeout/retry behavior validation.

  • Reliability and resilience checks including error recovery and fallback behavior.

• Share results and insights; recommend remediation and preventive actions.

Required Qualifications

• Experience

• • 5+ years in software QA, including test strategy, automation, and defect management.

  • 2+ years testing AI/ML or GenAI features, with hands-on evaluation design.

  • 4+ years testing ETL/data pipelines and data quality.

• Technical skills

• • Python: Strong proficiency building automated tests and tooling (PyTest, requests, pydantic or similar).

  • API testing: REST contract testing, schema validation, negative testing.

  • GenAI evaluation: crafting prompt suites, golden datasets, rubric-based scoring, and automated evaluation pipelines.

  • RAG testing: retrieval relevance, grounding validation, chunking/indexing verification, and embedding checks.

  • ETL/data quality: schema and constraint validation, reconciliation, lineage awareness, data profiling.

• Quality and governance

• • Understanding of LLM limitations and methods to detect/reduce hallucinations.

  • Safety and compliance testing including PII handling and prompt-injection resilience.

  • Strong analytical and debugging skills across services and data flows.

• Soft skills

• • Excellent written and verbal communication; ability to translate quality goals into measurable criteria.

  • Collaboration with AI engineers, data engineers, and product stakeholders.

  • Organized, detail-oriented, and outcomes-focused.

Nice to Have

• Experience with evaluation frameworks or tooling for LLMs and RAG quality measurement.

• Experience creating synthetic datasets to stress specific behaviors.