How can visual, workflow-based interfaces support how social sciences and humanities (SSH) scholars engage with large language models (LLMs)? SSH researchers increasingly use LLMs in interpretive inquiry to identify political stances in historical correspondence, compare thematic structures in literary texts, and examine how prompt design and model choice shape interpretive outcomes. Despite this uptake, little empirical work has examined how SSH scholars integrate and evaluate LLMs within disciplinary practice. Addressing this gap requires accessible tools for non-computing experts that move beyond high-level technical principles toward practice-based engagement (Mittelstadt, 2019).

This presentation reports on a user-centred study examining how visual, workflow-based experimentation can support reflective and methodologically grounded engagement with LLMs (Hartson, 2012). Formative interviews with SSH researchers identified everyday analytical tasks, evaluative practices, and uncertainty. These insights informed researcher personas and task scenarios (Blomquist & Arvola, 2002; Cooper, 2004), implemented in ChainForge, an open-source visual interface that enables users to compare LLM outputs across prompts and models. Developed by Dr. Ian Arawjo’s lab, ChainForge supports tasks from prompt engineering to auditing model behaviour for bias (Arawjo, 2024). Workflows were tested with SSH researchers using think-aloud methods to examine how participants evaluated model behaviour during use (Lewis, 1988).

During the presentation, we will demonstrate a ChainForge workflow alongside an open-access textbook documenting these methods. We will:

1.Load questions about the Indigenous Act Bill.

2.Generate prompt variants.

3.Run prompts across multiple LLM models.

4.Compare responses to ground-truth data in a CSV.

5.Score accuracy across prompts and models.

6.Visualize performance patterns at scale.

As AI systems increasingly mediate how human lives are recorded and narrated, the digital humanities field is confronted with questions of interpretation and representational authority. Personal data is predominantly structured to serve institutional, commercial, and administrative interests rather than individual understanding (boyd and Crawford 2012). In marketing, health, and platform economies, narratives generated by individuals are routinely analyzed to predict behavior, segment populations, and optimize decision-making, often without providing meaningful interpretive access to the data producers themselves. Datafication and interface design shape visibility, agency, and power, emphasizing that data representations are never neutral but encode institutional priorities (Drucker 2011; Manovich 2010).

Medical humanities is a field within digital humanities that examines how illness, care, and embodiment are experienced and narrated beyond clinical or institutional frameworks (Kleinman 1988; Charon 2006, 2001). Its contribution lies in demonstrating that encounters with illness are shaped by narrative, temporality, and meaning, and that these dimensions are central to how individuals understand their lives over time. Building on this tradition, the project draws on a corpus of publicly available social media data in which individuals narrate their experiences of illness and their encounters with medical systems over time.

This corpus is processed with natural language processing pipelines and large language models (LLMs) to extract thematic labels, temporal anchors, and event sequences from narrative text. Preprocessing includes cleaning, language detection, tokenization, lemmatization, and named-entity recognition, followed by topic modeling or LLM-assisted theme induction and temporal information extraction to align events and encounters along a timeline.

This digital demonstration presents an in-progress archive for Kerala mural paintings that experiments with how Indic aesthetic and ritual concepts can shape digital design. Rooted in Sanskrit śilp texts such as Chitrasūtra and Citralakṣaṇa, the project treats murals not just as images to be viewed, but as nodes in a living network of darśan, rasa, and temple pedagogy. The prototype questions what happens when these epistemic and devotional frameworks are translated into data models, metadata fields, and interface choices.

The demonstration will introduce a working prototype that includes high-resolution mural images, structured metadata (iconography, śāstric references, temple context), and experimental tags for aesthetic-emotional registers. It will walk participants through specific design decisions: how to record proportions and narrative sequencing, how to represent ritual and spatial context, and where temple protocols or community preferences limit full “openness.” By foregrounding these frictions, the project asks how Indic categories can unsettle standard assumptions about access, neutrality, and universality in digital heritage platforms.

The intended audience includes digital humanists, archivists, heritage professionals, and scholars of religion and art history interested in non-Western knowledge systems, community stewardship, and multimodal DH practice. The archive is at a prototype stage, with ongoing work on data structures, controlled vocabularies, and user testing. The demo invites feedback on both conceptual framing and technical implementation, positioning the project as a testbed for “Indic digital humanities” that takes the untranslatable as a design principle rather than a problem to be solved.

Hugh Thomson’s illustrations for the 1894 edition of Jane Austen’s Pride and Prejudice are instantly recognizable, particularly the peacock with the lavish swirling tail that adorns the front cover. Thomson was a well-known Victorian artist best known for his drawings for an 1891 edition of Elizabeth Gaskell’s novel Cranford. Whereas most illustrated novels of the period, including other editions of Pride and Prejudice, include only a few images depicting specific scenes from the story, Thomson’s edition is notable not only for the sheer number of illustrations but also because they present an interpretation of Austen’s novel, including events that happen beyond its pages. Critics tend to focus on the importance of Thomson’s edition to the cultural history of Pride and Prejudice and the history of Jane Austen, rather than on the illustrations themselves or their relationship to the text. By making Thomson’s illustrations available for close examination, the digital exhibit Hugh Thomson’s Illustrations for Pride and Prejudice makes the illustrations available to students, scholars, and interested readers available for study. It takes advantages of the tools available in Omeka and linked open data to enable users to examine various groupings of illustrations side by side in a way that is not possible by flipping through printed pages.

This digital & physical demonstration work invites participants to treat the “untranslatable” in Barbara Cassin’s sense as a hands-on problem rather than a concept. I present a transparent workflow that preserves the translation trail of centuries old cultural artefacts, restoring them to reproducible using digitally fabricated tools, and welcome visitors to participate by completing the final translation themselves through an analog (manual) print-making activity at the table.

The table setup includes a versioned project package. Visitors can toggle on the screen between the original scan, restoration layers, and proof impressions to see how small choices (stroke reconstruction, loss compensation, tonal stabilization) reshape legibility, iconographic reading, and the feel of an impression, even when the intent is “fidelity.”

Participants will be guided through a short manual printing process using the provided plates and simple materials. They will produce their own new impression of an 1200 year old work and can compare it immediately against other proofs and states. This micro-workshop makes visible what digitization and modelling cannot fully carry: small details like pressure, ink behaviour, relief depth, and the interpretive judgement embedded in every translational “clean-up.”

The project argues that Digital Humanities remediation is never neutral, and that pairing traceable digital files with hands-on, embodied reproduction can support protective access, teaching, and repatriation-adjacent return of usable surrogates to living communities while safeguarding fragile originals.