How can a health assistant earn trust when privacy, stigma, and misinformation shape what people are willing to ask?
A two-week deployment study of an expert-grounded WhatsApp chatbot for culturally sensitive menstrual-health education, published at the top venue in human-computer interaction.
Behind the research
My contribution
I helped co-design and build the assistant, then translated what we learned in the field into concrete requirements for safer, culturally appropriate health AI.
The study
30 co-design participants · 2-week field deployment
Are programs built by AI agents as reliable as release-quality human versions?
Research supported by NSF Trusted CI studying where AI coding agents succeed and fail at reimplementing real Linux utilities from documentation alone, advised by Prof. Barton P. Miller.
Behind the research
My contribution
I designed the multi-agent methodology, built a Go-based tool to analyze source reachability through Clang, DWARF, and the linker, and rebuilt a legacy fuzzer as a concurrent, resumable, checkpointed Go tool for faster testing.