Your choices are remembered on this device.
Portfolio
Publications, engagements, open-source projects and experience across multilingual AI, security and governance.
Research
Paper titles are shown in their original language of publication.
Second author · error analysis lead
Error analysis across 10+ LLMs including DeepSeek, Qwen, Gemma, Llama and Fanar — investigating dialectal bias, hallucinations, Clever Hans effects and gold-label inconsistencies.
Solo author · presenting researcher
Ensemble of AraBERT, AraModernBERT and CAMeLBERT with SAMER lexicon features for 19-level readability. Ranked 1st in the Constrained Track and 3rd in the Open Track.
Solo author · presenting researcher
Presenting how human traditions of listening and carrying memory can inform what we want from what and how we build.
Demos
Working prototypes from the Linguistic Security Institute, open on GitHub.
A prototype that probes LLM guardrails for "polyglot poisoning": using dialects, scripts and code-switching to slip past safety filters tuned mostly for English.
Responsible disclosure: full attack prompts are shared only with verified researchers working on defenses.
A decision tool for choosing which space-debris objects to take responsibility for first, built on seven questions grounded in the Lakota principles shared through the challenge.
Speaking
Talks and workshops across cybersecurity, linguistics, geoscience, education and space.
Pre-recorded virtual workshop plus a Saturday in-person, hands-on session.
How human traditions of listening and carrying memory can inform what we build.
A proposed framework for Indigenous data sovereignty in geoscience and education.
Hackathon contribution solving Challenge 2, presented via public showcase video.
Featured the Linguistic Firewall demo, examining vulnerabilities and defensive frameworks.
Parallels between human-language AI and non-human data ethics for space and conservation communities.
On the intersection of language, security and equity.
Open source
Python toolkit for detecting multilingual prompt injection, code-switching guardrail bypasses and polyglot poisoning. Presented at BSides San Diego 2026.
Interactive decision tool for responsible space-debris prioritization, built for the YOHAKU hackathon’s Challenge 2.
Retrieval-augmented text-to-speech prototype supporting Nahuatl-aware Spanish pronunciation, for Indigenous language preservation.
Early-stage command-line tool that flags hard-coded English and missing multilingual or dialect support in software projects.
Experience
Skills
English (native) · Spanish (native) · Arabic (reading, writing, linguistic analysis) · Italian (intermediate)
