// selected_work

Portfolio & case studies

Applied work spanning supply chain AI, human-in-the-loop evaluation, spatial & network modeling, and curriculum design — connecting fragmented data to decisions.

01Supply chain AI & information systems

case study

Mapping 107 seafood supply chains

Mapped 107 seafood supply chains across 50 stakeholder leaders — revealed that information asymmetry, not resource scarcity, drives coordination failure.

case study

Ceylon cinnamon, Sri Lanka

Analyzed the Ceylon cinnamon supply chain in Sri Lanka — revealed fragmented information, not production capacity, limits market access.

case study

ESG outputs for Fortune 500 clients

Evaluated ESG outputs for Fortune 500 clients (Mars, McDonald's, major retail and automotive brands) — revealed that sustainability-driven supply chain organization generates measurable financial returns.

02AI evaluation & NLP

human-in-the-loop

Precision gap in NLP pipelines (ONR / MURI)

Designed and benchmarked NLP pipelines for the Office of Naval Research: semi-automated methods achieved ~88% precision vs. ~38% for fully automated — quantifying the value of human-in-the-loop design.

eval

Conflict-event extraction, Sudan Tribune

Built an NLP/LLM pipeline processing all 2003–2010 news articles from the Sudan Tribune to extract and geolocate conflict events among 42 ethnic groups, enabling spatial analysis of environmental drivers of conflict.

eval

Failure-mode identification in production AI

Evaluated and iteratively refined AI outputs across supply-chain and analytical workflows; identified hallucinations and failure modes in production contexts before they reached decisions.

remote sensing

ML classification diagnostics for remote sensing

Developed and validated ML classifications for remote sensing; diagnosed systematic model errors that propagate downstream into environmental and land-use models.

03Spatial & network modeling

network science

Multi-layer behavioral networks, southern Chile

Built multi-layer behavioral networks linking 106 fishers, 311 species, 11 communities, and 17 management areas — revealing cross-scale interactions between land, ocean, and human behavior.

land cover

Land cover misclassification at scale

Validated machine learning land classifications at ~200 km local and ~1,200 km regional scales — revealed that misclassifying tree plantations as native forest distorts environmental models.

04Curriculum & AI literacy

curriculum

AI-assisted spatial analytics curriculum (ASU)

Designed an AI-assisted curriculum for Arizona State University's Institute for Social Science Research: 5 modules, 20 skills, 4 datasets — demonstrating that non-experts can evaluate spatial AI outputs with structured frameworks.