NER/RE Result of a Gemini 3.1 Flash Lite powered Extraction
13y of research meets 3 days of coding. NER/RE of plain text coin descriptions via Gemini 3.1 Flash Lite Batch API.
- 8k unique strings
- 99.4% syntactic integrity (JSON)
- Total cost: €8 (0.1¢/record)
- promising results, human review pending
- More on GitHub soon.
#DigitalHumanities #AI #NLP
Release: Agentic Reasoning Prototype!
Instead of forcing LLMs to hallucinate confidence, we provide a graph based ground truth (including ambiguity and conflicting interpretations).
github.com/imagines-num...
#Neo4j #MCP #ADK #GraphRAG #AI #OpenSource #KnowledgeGraph
Agentic Prototype demonstrating the IN.IDEA model: "Real" reasoning over Epistemic Graphs, instead of simulated doubt.
Powered by ADK, MCP, and Neo4j to map structured scholarly discourse directly.
resonism.substack.com/p/real-reaso...
#ADK #MCP #Neo4j #GraphRAG #DigitalHumanities #Numismatics
I was tired of treating AI as a carpenter, so I experimented with affective steering, using imageless "moods" to trigger universal resonance.
It’s not about hacks; it’s about the mechanics of meaning.
resonism.substack.com/p/semantic-i...
#DigitalHumanities #PromptEngineering
A deep ontology is a virtue, but recursion scales poorly.
We crave nuance in theory and performance in practice.
By trading recursive traversals for materialized paths, we achieve both without sacrifice.
resonism.substack.com/p/breaking-t...
#DigitalHumanities #GraphDB #Neo4j #RSEng
My current backlog: A physical manifestation of the "Great Unread." From the ethical imperatives of Hans Jonas to the technical ontologies of Gilbert Simondon, the stack is a reminder of why I need a navigational aid.
How to read philosophy when time is short? I’m testing a personal "Selective Reading" protocol: using LLMs before diving into the physical stack. Not an academic method, just my private attempt to stay engaged.
resonism.substack.com/p/selective-...
#DigitalHumanities #AI #Reading
(2/2) Applying our reified ‘interpretive act’ to such deep chains — without Barabucci’s ‘collapse to reality’ — would be a fascinating stress test for our graph’s performance. I think we should work through this case by developing a dedicated ‘Provenance Module’ within IDEA.
(1/2) Thank you for the pointer, Camilla. Barabucci et al. certainly put their finger on the 'data sins' we are trying to address.
While IDEA currently focuses on the structural analysis of visual content, their Colosseum example highlights the vital complexity of provenance and object biographies.
If we force genuinely ambiguous data into binary choices to fill our Knowledge Graphs, we risk an AI that has no choice but to claim facts where a human expert would express doubt.
Full read: resonism.substack.com/p/uncertaint...
#DigitalHumanities #AI #KnowledgeGraph
Thank you. The cold start is a sobering reminder of the task's scale. v0.9.2 is just a modest first step. We released it early precisely because we need the community's scrutiny to find the flaws in our logic. There is still so much to learn.
Diagram of the IDEA (Iconographical DefinitonAL Epistemological Architecture) graph model. It shows a central "Interpretation" node with attributes like "certainty: 0.7" and "reasoning_statement," linked to "CompositionEntity," "Agent," and "SourceReference." Crucially, it shows an "IMPAIRED_BY" relationship leading to "Architectonic Methodology" nodes for "Ambiguity" and "Source of Uncertainty," illustrating how the model formalizes epistemic doubt.
IDEA v0.9.2 is live.
Modeling just two synthetic fixtures took quite a while, the "cold start" in image reasoning is brutal.
We’ve built the tool's blueprint, join us in defining the standards of how to use it.
github.com/imagines-num...
#RSEng #GraphDB #Neo4j #DigitalHumanities #OpenScience #AI
Terminal screenshot showing Docker container logs for a successful deployment test. The logs detail the startup of the 'idea_neo4j_db' container and the completion of the 'idea_neo4j_seeder' container, which executed constraint, index, and synthetic fixture import scripts, ending with a green checkmark indicating "Seeding completed successfully" and exit code 0.
Tomorrow we release the blueprint v0.9 for one of our central tools, the foundation for our work over the coming decades.
More than a manual: we’ve included a demo with synthetic fixtures. Feel free to adapt the model or join us on the journey.
#RSEng #GraphDB #Neo4j #Numismatics #DigitalHumanties
Screenshot of the VS Code editor displaying the README.md of the 'IDEA' framework (Iconographical Definitional Epistemological Architecture). The sidebar shows a modular repository structure with folders for documentation, a domain-agnostic core, and a numismatics extension. The main text explains that IDEA bridges the gap between humanities research and high-performance data engineering, specifically addressing scalability issues of CIDOC CRM and the need for native modeling of scientific uncertainty and expert reasoning.
v0.9 of our IDEA (Iconographical Definitional Epistemological Architecture) is staged. Apache 2.0 license set.
Repo is populated. Final sanity check on Monday; aiming for Tuesday release. Quality over velocity.
#RSEng #GraphDB #Neo4j #AI #DigitalHumanities
A tabletop scene with a stone surface. On the left, a stack of books: Hans Jonas "Das Prinzip Verantwortung", Matthew B. Crawford "The World Beyond Your Head", and a large book about garden architecture "Karl Foerster. Der Staudenschöpfer". On the right, a children's book "Weihnachten mit Nulli & Priesemut" and an orange mug with a winking cartoon face ("Die Maus"). Pine branches and lights in the background.
Calibrations.
Wishing you clarity and focus. See you in the new year.
#RSEng #AI #DigitalHumanities
Core fixtures for the v0.9 graph model release in January 2026 are nearly finished. I’d hoped to wrap this up before the holidays, but I’m stopping at the 90% mark. I’d rather re-evaluate with fresh eyes in January than rush the final validation now.
Done for the year.
#RSEng #GraphDB #Numismatics
Screenshot of a presentation slide showing the noe4j labels for the compositional layer
RDF reification for complex image semantics often leads to structural bloat.
We use the Property Graph to keep Modifiers and Events as first-class citizens. No "triple-spaghetti," just maintainable logic and performance.
Logic first, serialization second.
#RSE #GraphDB #Neo4j #DigitalHumanities
Screenshot showing a SQL query to fetch free-text coin descriptions
Wrestling with mapping free-text coin descriptions to our graph. The NLP/RE pipeline is almost ready, but semantic ambiguity is a beast. Human language nuances strike back every time.
Definitely not a bash script job for the lunch break.
#RSEng #Numismatics #DigitalHumanities #NLP
Neo4j for the epistemic/query layer.
However, Postgres will be our Single Source of Truth. We love strict relational integrity for data management.
We treat the graph as a projection. Writing recursive CTEs is technically possible in SQL, but Cypher is just kinder to my sanity.
Arrows.app Screenshot showing reification of a tripple
Polishing the graph model before release.
Current struggle: Relations. RDF suggests "Hero -holding-> Sword". But our ontology demands Reification: "Hero -REL-> Holding -OBJ-> Sword" to properly model the act itself.
Praying I won't regret the extra hop cost later. 🤞
3/3 #GraphDB #Neo4j
Currently pivoting 50k legacy records from SQL to Graph using AI. Why? Because recursive self-joins to simulate ontology traversals only look good on paper. In production, they hurt.
2/3 #RSEng #Numismatics #DigitalHumanities
Breaking out of the cold start.
We are one year into our 25-year mission to build the Thesaurus of Ancient Greek Numismatics.
www.imagines-nummorum.eu/en
🧵 1/3 #RSEng #Numismatics #DigitalHumanities