I bring the precision and risk-discipline of aviation operations to the evaluation and training of large language models. My work centers on structured assessment of AI-generated outputs — ranking responses across accuracy, safety, and reasoning quality, identifying factual and logical errors, and providing the detailed rationale that supports RLHF and fine-tuning pipelines.
I help teams building AI products get the evaluation, rubric design, and quality assurance layer right. Before a model ships, someone needs to stress-test its outputs, catch the edge cases, and provide the structured feedback that actually improves performance — that's the work I do, day in and day out, across coding, text, and reasoning domains.
With hands-on experience evaluating and ranking outputs from leading LLMs, I understand what "good" looks like from the inside — and I can help you define it for your own product.
Start an AI project →Rubric authoring, head-to-head model comparison, and structured evaluation workflows to improve AI output quality.
Precise, adversarial, and domain-specific prompts designed to stress-test LLM reasoning and surface weaknesses before your users do.
Systematic review of AI-generated content for factual errors, logical fallacies, and misleading claims — across coding, math, and STEM domains.
Detailed written rationale and structured feedback that feeds directly into model improvement and fine-tuning pipelines.
A background in aviation systems and security — bringing safety-critical rigor and domain fluency to AI evaluation work in aerospace and related industries.
I start by fully internalizing the evaluation criteria and project specifications, so scoring is consistent from day one.
I assess AI-generated outputs across accuracy, safety, and reasoning — applying head-to-head comparisons where needed.
Every score comes with clear, written rationale — so my feedback is useful to model training, not just a number.
I cross-check against benchmarks and collaborate with other reviewers to maintain high inter-rater reliability before final submission.
I'm Salah Hassan. My path started in aviation. I studied Aeronautics at the University of North Dakota, where coursework in flight physiology, air traffic control, and aviation safety taught me how to think in systems: every component matters, every failure mode needs a plan, and precision isn't optional when the stakes are real.
That mindset carried directly into my current work: evaluating and training AI models. Since 2024, I've worked hands-on evaluating AI-generated outputs across code, text, and reasoning domains — ranking responses, writing structured rationale, and identifying the edge cases and errors that make the difference between a model that's good and one that's actually reliable.
I bring the same discipline from aviation security — where I managed compliance and risk processes for a major commercial airline — to every rubric I score and every prompt I write. I'm not chasing volume; I'm chasing accuracy.
Whether you have a detailed spec or just an idea, I'm happy to talk through your project and see if we're a good fit. I typically respond within 24 hours.
AI evaluation is the process of systematically assessing AI model outputs for quality, accuracy, reasoning, factual correctness, safety, instruction following, and overall usefulness using standardized evaluation criteria.
Prompt engineering is the practice of designing clear, structured, and effective prompts that guide AI models to produce accurate, reliable, and high-quality responses.
Prompt writing involves creating realistic user requests that effectively test an AI model's reasoning, creativity, instruction following, coding ability, factual knowledge, or problem-solving capabilities.
Evaluation rubrics are structured scoring frameworks used to assess AI responses across multiple dimensions such as accuracy, completeness, reasoning, clarity, safety, instruction adherence, and overall quality.
Even advanced AI models can produce inaccurate, biased, or incomplete responses. AI evaluation helps identify these issues and provides valuable feedback that improves model performance and reliability.
I evaluate responses side-by-side using predefined rubrics, comparing reasoning quality, factual accuracy, completeness, instruction following, efficiency, safety, and overall user experience.
A strong prompt is clear, specific, context-rich, unambiguous, and designed to encourage the AI to produce accurate, complete, and relevant responses.
Worked with me or just visited? I'd love to hear what you think. Your feedback helps me improve and grow.