Rasa Rasiulytė

Exploring practical approaches to evaluating AI systems, with an emphasis on code quality, safety, and reliability.

• Evaluating LLM-generated code at Outlier.ai, identifying correctness, safety, and quality issues
• Studying and applying AI evaluation approaches such as LLM-as-Judge, Process Reward Models, and multi-agent frameworks
• Developing familiarity with governance frameworks (NIST AI RMF, EU AI Act, ISO/IEC 42001) and thinking through how high-level requirements translate into testable criteria
• Investigating which traditional QA skills transfer well to evaluating non-deterministic systems, and where new approaches are needed

Xbox Backward Compatibility
Worked on infrastructure to ensure legacy games continued to function correctly on new hardware.

• Built validation pipelines to detect compatibility issues before release
• Collaborated closely with hardware, platform, and testing teams to diagnose and prevent regressions

Office Security
Performed security testing and fuzzing for Microsoft Publisher.

• Designed fuzzing approaches that uncovered buffer overflows and security vulnerabilities

Windows Movie Maker / Expression Encoder
Conducted exploratory testing focused on encoding, graphics compatibility, and edge cases.

• Identified critical rendering and playback issues across a wide range of hardware configurations

Notes and short essays I've written while working through questions around AI evaluation and software quality:

• "What I Look For When Evaluating Code" — How I think about code quality and recurring patterns
• "From SDET to AI Evaluator: What Transfers?" — Which testing instincts still apply to AI evaluation, and which don't