TDWG2026
Home / Talks / SAL A
talk · Friday 25 September · 10:42–⁠10:54 · SAL A

Contained Agentic Workflow for Literature Analysis and Data Extraction in Museum Collections

Steen Dupont · Large Language Models for Biodiversity Data Discovery, Integration, and Curation. AI applications, integration, and validation

Recording time 3:13:32–3:24:40Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionContained, local agentic RAG over a literature corpus is feasible on a laptop or shared edge machine, avoiding the security worries of cloud agents in museums.

Overview

What this was about

Steen Dupont demonstrated Kvasir, a prototype built mostly by coding agents (Codex) that runs fully on local hardware. It ingests a personal literature corpus with Docling, stores it in Neo4j and Qdrant, and lets an OpenClaw agent with a local Qwen model answer questions and explore papers. The demo ingested papers into 696 chunks and showed papers broken into sections with their references plotted over time, to characterise papers and trace evidence across the corpus. He argued for local AI for security, noted that optimised local models are improving, and suggested shared edge machines (e.g. a Dell GB10 serving 10-20 users) as institutional infrastructure, with the agent as a pluggable orchestration layer.

Why it matters. Museums worry about releasing agents on institutional data. This prototype shows a low-cost, offline route to experimenting with LLM literature analysis.

Key ideas

In the room

  • The project explores AI on local hardware rather than cloud LLMs; museum IT set up a closed network, but the demo runs offline on a laptop.
  • Motivation: better visibility of a PhD literature corpus: most-cited works, citation timing, references per section, without reading every paper.
  • Architecture: OpenClaw agent orchestrator, Qwen 3.6 main model, Docling parsing, Codex-written Python scripts, Neo4j graph and Qdrant vector store, Kokoro for audio.
  • Demo: two papers ingested, 696 chunks created, a paper view with sections, references per section and their publication years.
  • Idea: the distribution of references (introduction-heavy vs conclusion-heavy) characterises paper type.
  • Why local AI: security, improving local models, and pre-processing so the LLM only does the important step.
  • A GB10-class edge machine could serve 10-20 users on ordinary laptops instead of expensive individual machines.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 22 Sept · 11:30–⁠11:45 · SAL B

From Images to Structured Data: A Scalable AI Workflow for Natural History Museums

A simple 'just do it' AI transcription pipeline with schema prompts, an intent step and full provenance can make label digitisation scale.

Anne Koivunen
talk · 22 Sept · 16:05–⁠16:10 · SAL B

Agentic AI Systems for Specimen Interaction and Camera-Based Observation

Off-the-shelf local vision language models can pick out and count specimen morphology and spot damage with minimal guidance.

Jack Hollister
talk · 25 Sept · 11:06–⁠11:18 · SAL A

Integrating Many Data Sources in iChatBio Enables Emergent Scientific Use Cases

iChatBio's specialised agents let a conversational interface combine many biodiversity data sources, with traceable, non-hallucinated data artefacts.

Michael Elliott
Programme abstract

Contained Agentic Workflow for Literature Analysis and Data Extraction in Museum Collections

Steen Dupont

As agentic AI systems become more capable, they offer new opportunities for interacting with complex digital infrastructures, specialist literature and collections data. These systems can move beyond single-purpose language models or fixed Retrieval-Augmented Generation workflows by supporting more flexible, context-aware interactions with knowledge. However, highly versatile agentic systems also raise important concerns around cyber security, governance, compute requirements and token costs, particularly when deployed at institutional scale.

This presentation explores the use of local LLMs and agentic workflows for the analysis of scientific literature and the extraction of structured data relevant to research and museum collections. Operating on local hardware, the system is designed to ingest, interpret and curate knowledge from research papers, reports and archival documents, linking extracted information to existing collections records and domain knowledge.

Using Dell GB10 systems and high-performance devices, we demonstrate how contained LLM workflows can identify taxonomic, morphological, geographic and ecological information within text, converting narrative scientific knowledge into structured, machine-readable data. This enables the enrichment of collections databases and supports the integration of external knowledge sources into digitisation workflows.

By operating within a local/on-device environment, the approach maintains control over data processing. Live demonstrations will show how local LLMs and agentic systems can augment research and curation workflows by helping researchers and curators unlock information contained within scientific literature and integrate it directly into research and collections-based work.

Submitted to the conference programme before the talk, so the talk itself may differ.