Jon A. Chun

Software

Research code, benchmarks and teaching material, all public on GitHub and maintained since 2019. The methods below were developed to make humanistic questions measurable without flattening the ambiguity that makes them interesting, and to build evaluation into systems from the start rather than bolting it on at review time.

  • SentimentArcs

    A large ensemble of dozens of sentiment analysis models for analyzing emotion in text over time. Model disagreement is treated as interpretive data, curves are downsampled with LTTB to preserve peaks and valleys, and dynamic time warping compares arcs across texts of unequal length. The technical basis for The Shapes of Stories (Cambridge University Press, 2022).

    Jupyter Notebook 43 stars 10 forks

  • MultiSentimentArcs

    A novel method to visualize multimodal AI sentiment arcs in long-form narratives, and the first fully open-source framework for diachronic multimodal sentiment analysis. Measures emotional coherence across film dialogue and imagery.

    Python 6 stars 4 forks

  • AIStorySimilarity

    A benchmark for long-text story similarity grounded in narrative theory, with applications to search, intellectual property infringement and guided creativity. Published at ACL EMNLP/CoNLL 2024.

    Python

  • LLM Ethics Audit

    A novel ethics-based audit of the state-of-the-art LLM chatbots, comparing the ethical frameworks the leading models invoke and where those frameworks diverge under pressure.

    Audit framework 1 star

  • GenAI Multi-Agent Networks and Digital Twins

    Generative AI, multi-agent systems, AI research methodology, industry best practices and the future of work. The full course repository for Kenyon College IPHS 391, Fall 2025.

    Jupyter Notebook 24 stars 7 forks

  • IIoT Time Series Prediction System

    An end-to-end industrial Internet of Things time series prediction system, built for predictive maintenance and developed with an industry collaboration.

    Python 4 stars

  • CEI Benchmark

    The Contextual Emotional Inference benchmark: a pragmatic reasoning dataset of 300 human-validated scenarios with replication code. Data under CC BY 4.0, code under MIT.

    Python

  • AgenticSimLaw

    Multi-agent debate simulation of a US bench trial predicting juvenile recidivism, with prosecutor, defense and judge roles over a seven-turn structured debate. Accepted at the AAAI-26 LaMAS workshop.

    Python

  • International AI Regulation Survey

    Code and data comparing AI regulations in the EU, the US and China, supporting the comparative global AI regulation work.

    Python

Block diagram of the SentimentArcs pipeline, showing the flow from text ingestion through the ensemble of sentiment models to smoothing, downsampling and trajectory comparison
Block diagram for the SentimentArcs notebooks.

All public repositories

Every public, non-archived repository, grouped by theme. Scaffolding and empty repositories are omitted. Regenerate this list with python3 scripts/build_repos.py.

SentimentArcs & Narrative 63

AI Evaluation, Safety & Policy 13

Agents & LLM Tooling 18

Courses & Teaching 26

Data Collection & Scraping 14

Dev Tooling & Misc 39