Payoffs of Haystack Tagging: AI Ready

Tagging and data modeling are transforming how the building industry manages information, and Project Haystack is at the forefront. In our blog series, "The Strategy and Payoffs of Tagging," we share practical examples of how tagging and data modeling are being used by the Haystack community in a variety of applications and use cases. In this blog, we focus on how tagging paves the way for making building data AI-ready.

AI is becoming a new frontier for consuming and optimizing intelligent building data designed to harness the power of large language models (LLM) and generative AI to perform intelligent and complex tasks. 

Generative AI focuses on creating new content, like responses to questions using natural language, by learning from existing data. Predictive AI, on the other hand, analyzes historical data to forecast future outcomes or optimizations.

This new technology works best with large data sets by detecting patterns, when applied to smart buildings with smaller data sets and higher complexity there is a need for structured data. An emerging trend in our industry is to have an independent data layer that abstracts the siloed subsystems of unstructured data into a common format. (native structured data)

The fundamental problem with LLMs and small data sets is that they are more likely to hallucinate (provide false information or conclusions). Adding context and metadata to your building's information is crucial. The sensor data produced by control systems is raw and typically unformatted. As Nexus Labs put it,"garbage in, garbage out." 

For example, a future intelligent assistant would be prompted with a natural language question: "How is my building's energy performance this week versus last week?" Behind the scenes, the AI would convert that prompt to something that is machine-readable, in the form of a Haystack query. The structured data for historical energy usage would be easy to identify using tags, such as elec + meter combined with the timeframe for date, which would gather last week vs this week's electricity usage. The AI would then compare mathematically some of the energy usage and report back in a natural language response, the results. 

AI could also be used for querying if lights are left on, providing a summary of alarms, and intelligent equipment optimization (to name a few). Here's a great resource for more real-world examples of how AI and data tagging are making smart buildings more efficient, reliable, and comfortable.