In June 2025, a local LLM was constructed based on Franklin and Sterling Hill mineralogical data. The purpose of this model is to assist with the identification of Franklin-Sterling Hill minerals.
The model is called “Franklinium””. It is a retrieval-augmented generator (RAG) with BM25 tuned for mineral terms and symbols . The model is physically hosted on a custom Intel i7 with an Nvidia RTX5060 GPU running Windows 11. The model is isolated and is not part of public GPT5, Gemini, Copilot, or other LLMs.
As of August 2025, Franklinium is trained on the following sources:
- Descriptions and morphology of over 375 Franklin-Sterling Hill minerals
- Fluorescence data (shortwave, midwave, longwave)
- Raman spectra (785nm-1064nm)
- Published historical data from Albanese, Palache, Frondel, and Dunn
- FOMS Picking Table journals
The Franklinium LLM will be made available to researchers in the new future. Please contact us if you would like to participate in this project and supply data!