fancy (research) tools!Agentic engineering for everyone who works with knowledge.

Tools to adapt

• LLM-assisted
• Expert review in the tool
• Overview of the whole collection
• Readable text instead of tags
• Runs in the browser without a server of its own
• Evidence back to the source

• 

  coOCR/HTR

  Recognises the text of early printed books and manuscripts and lets every reading be checked against the page image.

  Research preview

  • LLM-assisted, model of your choice up to a local model
  • Expert review in the tool
  • Runs in the browser without a server of its own

  Open tool Code Evidence Request adaptation

• 

  teiCrafter

  Edits TEI XML as readable text and validates against the schema on saving.

  Research preview

  • LLM-assisted, model of your choice up to a local model
  • Readable text instead of tags
  • Expert review in the tool
  • Runs in the browser without a server of its own

  Open tool Code Evidence Request adaptation

• 

  CorrespExplorer

  Shows letter metadata from CMIF as a map, a timeline and a network, so that it becomes visible who wrote to whom, when and from where.

  Live demo

  • Overview of the whole collection
  • Runs in the browser without a server of its own

  Open tool Code Evidence Request adaptation

• 

  SZD-HTR

  Transcribes an entire estate and records for every object whether a machine, an agent or a human has checked it.

  Experiment

  • LLM-assisted
  • Overview of the whole collection
  • Expert review in the tool

  Open tool Code Evidence Request adaptation

• 

  Stefan Zweig Bibliography

  Rescues a bibliography from a decommissioned wiki into research data in which every statement keeps its source page.

  Holdings secured

  • LLM-assisted
  • Overview of the whole collection
  • Evidence back to the source

  Open tool Code Evidence Request adaptation

Principles

• Project knowledge and method

  Work begins with an understanding of the project, its research question, its data and its workflows. This knowledge is recorded in a project knowledge base and versioned with the code. The method of context engineering follows from it, for example Promptotyping, Grounded Vault, an agentic pipeline or a simple work cycle with one agent.

• A framework for coding agents

  The quality of agent-generated code depends on the context in which the agents work. We design this framework from the knowledge base, precise requirements, sample data from your holdings and automated tests that the agents run themselves.

• Declared maturity

  Results are prototypes and research tools whose maturity is openly declared. Productive operation, for example with sensitive data or many users, requires professional revision and an independent review of the code.

• Open formats

  Results are held in open, documented formats, depending on the material TEI, PAGE XML, METS/MODS, JSON-LD or CSV. Image data can be integrated via IIIF. The data thus remain readable independently of the tool and can be transferred to other systems and repositories.

• Expert control

  Where large language models are used, it remains discernible for every item whether it was generated by a model, checked by an agent or confirmed by your experts. Only content confirmed by your experts counts as established.

• Handover

  You receive the complete source code and the project knowledge base, on the basis of which your team, external developers or AI assistants can continue the work. For research data we offer long-term archiving in the certified repository GAMS.