03/23/2026
Prototype for AI reasoning in numismatics
New Github repository published: GraphRAG-Driven Agentic Reasoning: A Research Prototype using ADK, MCP, and Neo4j
Over the last 2 weeks, a functional prototype for Agentic Reasoning was developed to showcase our IN.IDEA Graph Framework for Imagines Nummorum in action. The "Curator" demonstrates how structured research data can be used as "External Ground Truth" for AI agents.
The Stack:
- GraphDB: Neo4j (+ Vector Index)
- Protocol: MCP (Model Context Protocol)
- Agent: ADK (Google Agent Development Kit) Gemini 3.1 Flash Light Preview (Reasoning doesn't require a flagship model when the graph does the heavy lifting)
Why this matters:
Instead of forcing the LLM to simulate discourse and doubt, we provide it with the data (including all its uncertainties and diverging opinions). It doesn't act as a surrogate scientist, but as a data curator that is comprehensible to humans—unlike the mathematics of AI latent space.
Repository: github.com/imagines-nummorum/agentic-graph-rag
For a quick overview without installation, you will find
here a short introduction, including Screenshots.
Автор: Jan Köster
Назад к обзору новостей
