WORK
Human-Centered AI & Healthcare Research
LUMS
Used the Anthropic Claude API to generate and evaluate 40+ LLM healthcare responses across socially and culturally sensitive user scenarios, designing prompt structures and collecting model outputs for systematic analysis. Built a structured evaluation framework to assess responses across 9 dimensions, including accuracy, bias, cultural sensitivity, misinformation, safety, trust, usefulness, contextual fit, and guardrail performance. Conducted 22 semi-structured interviews and collected survey responses from 100+ participants to compare how clinicians and non-clinicians evaluate AI-generated health advice in culturally conservative contexts. Used ATLAS.ti for thematic coding and qualitative analysis, synthesizing interview data, prompt-response patterns, and evaluation scores into recommendations for safer Human-Centered AI deployment.
•Used the Anthropic Claude API to generate and evaluate 40+ LLM healthcare responses across socially and culturally sensitive user scenarios, designing prompt structures and collecting model outputs for systematic analysis.
•Built a structured evaluation framework to assess responses across 9 dimensions, including accuracy, bias, cultural sensitivity, misinformation, safety, trust, usefulness, contextual fit, and guardrail performance.
•Conducted 22 semi-structured interviews and collected survey responses from 100+ participants to compare how clinicians and non-clinicians evaluate AI-generated health advice in culturally conservative contexts.
•Used ATLAS.ti for thematic coding and qualitative analysis, synthesizing interview data, prompt-response patterns, and evaluation scores into recommendations for safer Human-Centered AI deployment.
Key Achievements
→Discovered that both clinicians and non-clinicians utilize AI for health advice in conservative contexts, highlighting a critical need for professional-grade AI guardrails.
→Conducted 22 semi-structured interviews to evaluate AI-generated health advice across 9 dimensions of accuracy and cultural sensitivity.
Anthropic Claude APILLMATLAS.tiQualitative AnalysisThematic CodingPrompt Engineering