At Cercana, we’re excited by the constant evolution of geospatial technology. AI and its related technologies are becoming increasingly important components of geospatial workflows. Recently, our founder, Bill Dollins, has shared some of his explorations into AI through his personal blog, geoMusings, where he has written about topics like Named Entity Recognition (NER), image similarity using pgVector, and retrieval-augmented generation (RAG). These explorations reflect Cercana’s commitment to helping our clients understand emerging technologies and consider how best to integrate them into their operations.
In his May 2024 post, Bill explored the application of ChatGPT for Named Entity Recognition (NER), a vital tool in the AI toolkit. NER can extract key information from unstructured text, such as identifying people, locations, and organizations. At Cercana Systems, we see potential in using AI to streamline geospatial data processing, particularly in the context of large-scale data integration tasks. By using AI tools like ChatGPT, we can automate the extraction of spatial information from textual data, making it easier for our clients to analyze and take action.
This capability is particularly relevant in scenarios where large volumes of data need to be sifted through quickly—whether for real-time monitoring or in-depth analysis. As we continue to refine our capabilities, Cercana is positioned to offer more precise and scalable solutions.
Exploring Image Similarity with pgVector
In a July 2024 post, Bill explored analyzing image similarity using pgVector, a vector database extension for PostgreSQL. This post examines the creation of direct vector embeddings from images, rather than solely using image metadata such as EXIF tags. Combined with such metadata, including location, direct embeddings enable a more discreet kind of “looks like” analysis on a corpus of images.
By integrating pgVector with existing geospatial data pipelines, we are enhancing our ability to process and analyze visual data more efficiently. This capability not only speeds up workflows but also opens new avenues for our clients to derive actionable insights from their image datasets.
Experimenting with RAG Using ChatGPT and DuckDuckGo
Most recently, in August 2024, Bill explored the concept of retrieval-augmented generation (RAG) by combining ChatGPT with DuckDuckGo for information retrieval. RAG models are a quickly-developing capability in AI because they blend the generative capabilities of models like LLMs with the precision of traditional search engines and database queries. This fusion enables more accurate and contextually relevant information retrieval, which can be valuable for analytical tasks and other data-intensive operations.
At Cercana, we’re looking at RAG to enable AI-enhanced geospatial solutions. By integrating RAG, we seek to provide clients with more discreet, context-aware tools that can make sense out of large volumes of unstructured data.
Moving Forward
As Cercana grows and evolves, we are finding new ways of integrating AI into our geospatial services. Bill’s explorations into NER, image similarity, and RAG are examples of this. We believe that AI tools and related technologies, when paired with more traditional tools such as data pipelines and geospatial databases, provide the opportunity to improve data quality and shorten the time-to-market for actionable information.
To learn more about how Cercana can help you integrate AI into your operations, please contact us.