Technical Expertise
Building AI systems that ship
AI & ML: Full-stack development from GPT-2 through contemporary LLMs (GPT-4, Claude, Llama, Mistral). Python, TensorFlow, PyTorch, scikit-learn, HuggingFace Transformers. LLM red/blue team testing, threat modeling, vulnerability detection, XAI, uncertainty quantification. Created SentimentArcs (open-source, adopted globally). Multi-agent architecture design, game theory simulations, automated agent frameworks, multimodal AI.
Earlier career: C/Assembly/UNIX systems programming (Lawrence Berkeley Labs). Enterprise security architecture. Two US patents on network security. Japanese semiconductor patent analysis (SEMATECH). Medical informatics (AHA Research Fellow).
Earlier career: C/Assembly/UNIX systems programming (Lawrence Berkeley Labs). Enterprise security architecture. Two US patents on network security. Japanese semiconductor patent analysis (SEMATECH). Medical informatics (AHA Research Fellow).
Research Projects
What does Jon Chun research?
NIST AI Safety
LLM Security Evaluation & Red-Teaming
PI representing the 25,000-member Modern Language Association at the US AI Safety Institute Consortium (CAISI). Focus: how language models process negation, prohibition, and persuasion. Evaluating robustness of ethical guardrails across 16 models and 14 adversarial scenarios.
Schmidt Sciences
Archival Intelligence
PI for $330K Schmidt Sciences HAVI grant. One of 23 teams worldwide. Building AI tools for endangered cultural archives in New Orleans — automated transcription, entity extraction, and semantic search for handwritten historical documents.
Multi-Agent Systems
Behavioral Simulation & Game Theory
Multi-agent debate simulations for high-stakes decision making. AgenticSimLaw: juvenile courtroom simulation accepted at LaMAS 2026. Exploring emergent behavior in adversarial agent populations with Notre Dame and IBM.
Computational Methods
SentimentArcs & Narrative AI
Created SentimentArcs, an open-source methodology for diachronic sentiment analysis of narratives. 95,000+ downloads from 4,000+ institutions across 198 countries. Extended to multimodal analysis (text, audio, video) in MultiSentimentArcs.
What is SentimentArcs?
SentimentArcs is an open-source computational methodology for analyzing sentiment trajectories across narrative texts. It applies ensemble NLP methods to map emotional arcs in literature, film scripts, and other narrative forms. Adopted globally with over 95,000 downloads from 4,000+ institutions in 198 countries, it has been used in 300+ student research projects spanning computational humanities, digital scholarship, and affective AI. The methodology has been extended to multimodal analysis through MultiSentimentArcs, published in Frontiers in Computer Science (2024).
Last updated