本文介绍了一款名为GEOAssist的开源地质AI应用，它利用大型语言模型（LLM）整合非结构化和结构化数据，实现自然语言查询。该应用能够检索参考文献，可视化板块构造重建，并集成Macrostrat和Gplates API，方便用户获取地质信息。最新版本V1.0增加了自主代理功能，可通过DuckDuckGo进行互联网搜索，自动下载和重命名PDF文献，并进行内容提取，极大地提升了地质文献管理和研究的效率。该应用旨在提升地质科学领域对AI的认知和应用能力，并鼓励用户利用开源数据和本地化部署，安全地进行研究。

🌍 GEOAssist是一款开源地质AI应用，核心功能是通过大型语言模型（LLM）实现对多源地质数据的自然语言查询。它能够整合非结构化和结构化的信息，如参考文献和可视化板块重建，极大地简化了地质学研究中的信息收集和综合过程。

🚀 应用集成了多种强大的外部API，包括来自Crossref和arxiv的参考文献数据，以及AuScope的Gplates API用于板块重建，和xDD University of Wisconsin-Madison的Macrostrat API用于地层单位信息。这使得用户能够便捷地访问和利用现有地质数据资源。

💡 GEOAssist V1.0版本引入了自主代理功能，能够通过DuckDuckGo进行互联网搜索，并自动下载、阅读PDF文件，提取关键信息以进行“自适应搜索”。此外，它还能自动重命名PDF文件为“作者_日期_标题”的格式，有助于地质文献的管理和浏览。

📚 该应用通过“自适应搜索”能力，能够处理复杂的查询，例如“讨论埃及的侏罗纪”，并自主发现、下载大量相关PDF论文和DOI参考文献。它还能生成通用AI摘要，并与Macrostrat和Gplates等数据源进行交叉验证，提供全面的研究支持。

🔒 GEOAssist强调了在Python中本地运行AI应用的可能性，用户可以使用自己的私有数据，并确保数据安全和隐私。这为研究人员提供了一个灵活且安全的解决方案，尤其适用于处理敏感或保密的地质信息。

Iniital 19th August 2025 Post:

I created an Open-source Geological AI App yesterday that lists reference literature and visualises plate reconstructions if relevant to the question. It was just to illustrate how we can use Large Language Models (LLM) to enable natural language querying of multiple sources, both unstructured and structured, to help us gather and synthesize disparate information together.

It is very easy to use off the shelf LLMs that already understand much geological terminology. I used the open Gemma model here as an example, you can either download an LLM and run locally for total privacy, or use a free Huggingface inference endpoint. The references are from Crossref, arxiv and others. The plate reconstruction is using the excellent hashtag#Gplates API from AuScope, you can choose which reconstruction model to use. The Lithostratigraphic Units table is generated using the also excellent, Macrostrat API from xDD University of Wisconsin-Madison

There is a lot of open-source data out there and it can be quite easy to develop your own AI apps in Python that run locally. You can use your own private data as well and keep everything local and secure.

I’ll add a few more things next weekend, clean up the code and share shortly. You can watch a video of the app and plate reconstructions here: https://www.linkedin.com/posts/paulhcleverley_gplates-geology-geoscience-activity-7363126587233370113-uw5Q?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAJdxjgBSnNfsxghi8atnlNooAgz4mP6AtE

Update: 21st August 2025:

GEOAssist: Opensource Geological AI App V1.0 released! Added an autonomous agent using DuckDuckGo for Internet search, downloading PDFs, machine reading them for additional terms for ‘adaptive search’.

All PDFs are saved to your downloads folder automatically so you can easily collect hundreds or more of files for a given query. PDF filenames are not typically helpful, so are also automatically renamed to the author_date_title leveraging text from inside the PDF, to aid geoscience literature management and browsing.

You can control the number of adaptive rounds; for example, a query “Discuss the Jurassic of Egypt” using GEOAssist, autonomously found and downloaded over 200 relevant PDF papers, several hundred DOI references, a Gen AI summary as well as checking Macrostrat and Gplates for relevant data. I posted about some of this last weekend.

The purpose is to raise awareness and capabilities in the geological sciences around AI. How easy it is to build your own AI apps, the sorts of things you can do, and open-source data that exists. It is easy to add your own private local sources and keep things secure for confidential content.

Code to run the app V1.0 is in Github. I will be updating periodically. Link to Github code below:

https://github.com/PCleverleyGeol/GeoAssist—An-open-source-autonomous-research-agent-for-geoscience-data-and-literature.-