Ai Building Energy Management | AI Final Year Project Report

Complete Technical Paper Content — Ai Building Energy Management

1. Executive Abstract, Problem Formulation & Background

An AI agent, also known as agentic AI, is an artificial intelligence program that can make decisions, pursue goals, and take actions independently of human programmers. Agentic AI contrasts with tool-like AI use for narrow, specific tasks such as answering questions, as with chatbots in the non-agentic forms that were common since 2023, or with traditional machine learning algorithms. While there is no universally agreed-upon definition of an AI agent, common attributes of AI agents include goal-directed behavior, use of external tools, the ability to interact with and modify an external environment, and the ability to autonomously perform multi-step tasks. Their control flow is frequently driven by large language models (LLMs). Agent systems may also include memory components, planning logic, tool interfaces, and orchestration software for coordinating agent components.

The theoretical underpinnings for artificial agents emerged in the mid 20th century, with establishment of cybernetics and artificial intelligence. Oliver Selfridge's 1958 "Pandemonium: A Paradigm for Learning" paper was an important early theoretical contribution in establishing agent oriented architecture. Practical implementations of agents for real world applications began to become widespread in the 1990s, after the introduction of the belief–desire–intention software model (BDI), and agent-oriented programming. Harvard professor Milind Tambe notes that in the 1990s, the definition of an AI agent was not clear. Pure digital agents were deployed in computer infrastructure for purposes such as monitoring, while agents connected to real-world sensors and actuators were increasingly used in industrial control systems. Early artificial agents tended to have simple if then logic, which expanded over time into large decision tree models. By the early 2010s, products like Siri and Alexa were released, and were sometimes called AI agents, though they lacked the general purpose reasoning ability of later agents run by large language models (LLMs). The development of LLMs during the late 2010s and early 2020s introduced new approaches to AI agents. Models such as GPT-3 demonstrated the ability to generate and understand natural language at a much larger scale than many earlier language systems. Academics began to study LLM agents from 2018. Deployment of such agents began to accelerate in late 2023 after OpenAI's "function-calling" API was made available, and especially after Anthropic's late 2024 introduction of Model Context Protocol (MCP), a standardised way for LLM agents to gain contextual awareness, and to act on the world by calling various external tools. The term "agentic" only began to be used with any frequency in 2024, popularised in part by researcher Andrew Ng.

Ken Huang proposed an AI agent reference architecture that consists of seven interconnected layers, with each layer building on the functionality of the layers beneath it:

Layer 1: Foundation models – provide the datasets that power the agent. Layer 2: Data operations – manages the data infrastructure required for AI agent operations, including vector databases, data loaders, and RAG. Layer 3: Agent frameworks – software that manages the AI agents. Layer 4: Deployment and infrastructure – the technical foundation of the AI agents. Layer 5: Evaluation and observability – the safety and performance of AI agents. Layer 6: Security and compliance – a protective framework for safe operation and compliance with regulatory boundaries. At this layer, security and compliance features embedded into all the AI agent stack layers are integrated together. Layer 7: Agent ecosystem – represents the AI agents' interface with real-world applications and users.

An agent harness, or agent scaffold, is the software layer surrounding a large language model that enables it to function as an AI agent. It commonly manages prompts, context, tool use, memory, execution state, operational constraints, sandboxes, permissions, and the processing of results. The harness connects the model to internal and external computer hardware, software, data files, databases, web browsers, command-line interfaces, and application programming interfaces, while controlling how the agent accesses and uses these resources to complete multi-step tasks.

2. Related Work, Literature Survey & Algorithmic Foundations

Prompt chaining: A sequence where the output of one step serves as the input for the next. Routing: Directing an input to a specialized downstream task or tool. Parallelization: The simultaneous execution of multiple tasks. Sequential processing: A fixed, linear progression of tasks through a predefined pipeline. Planner-critic: An iterative pattern where one agent generates a proposal and another evaluates it to provide feedback for refinement.

STRIDE: A Microsoft model that identifies threats across six categories: spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege. MITRE ATLAS: A knowledge base of adversary tactics and techniques for AI systems. OWASP GenAI Security Project: Provides guidance on vulnerabilities associated with generative AI and large language model integration, including guidelines specifically for agentic applications. MAESTRO: A framework from the Cloud Security Alliance for assessing risks in AI systems throughout their lifecycle.

The AI boom is a period of rapid growth in the field of artificial intelligence beginning in the mid-2010s and accelerating in the 2020s. Generative AI technologies include large language models (LLM), which generate text and code, image models, and video models, and world models. This period is sometimes referred to as an AI spring, a term used to contrast it with previous AI winters. Key advances preceding the boom include the transformer (2017), generative pre-trained transformer (2018) and reinforcement learning from human feedback. During the boom, LLMs have evolved via the reasoning model, mixture of experts technique, and multimodal learning. As of 2026, leading proprietary LLMs have been released by Anthropic, OpenAI, and divisions of Google, Meta, and SpaceX in the United States. In China, companies favour open weights releases, led by Alibaba Cloud, DeepSeek, Moonshot AI, and Z.ai. Controversies of the AI boom include its environmental impact, dominance by US Big Tech companies, effect on copyright, proliferation of deepfakes, threats to cybersecurity and biosecurity, its use in warfare, a potential economic bubble, and existential risks such as AI takeover resulting from artificial general intelligence and superintelligence. Since 2018, observers including Ian Bremmer, Nicholas Thompson, and Kai-Fu Lee have argued an AI arms race or AI Cold War exists between the US and China. The boom has impacted the semiconductor industry, causing a global memory supply shortage; As of second quarter 2026, Nvidia is the world's largest corporation by capitalization while TSMC, Broadcom, and Samsung are among the ten largest. ChatGPT is the fifth-most-visited website globally.

In 1950, Alan Turing proposed the idea of "Thinking Machines". These were computers that would be able to reason at the same level as humans. He began his well-known "Turing test", where an interrogator is provided with two materials and they must determine which one was done by artificial intelligence and which one was done by a human being.

In 1956, John McCarthy used the term "artificial intelligence" for the first time. That year, McCarthy, Nathaniel Rochester, Marvin Minsky, and Claude Shannon organized the Dartmouth conference, which formalized artificial intelligence as an academic field. In 1958, McCarthy created the programming language LISP, LISP stands for "List Processing" and was the main programming language for artificial intelligence. which remained the most common programming language for artificial intelligence in the United States for decades. In 1962, McCarthy founded Stanford Artificial Intelligence Laboratory (SAIL). McCarthy was also a cofounder of MIT's first Artificial Intelligence Laboratory, now known as MIT Computer Science and Artificial Intelligence Laboratory. In 1966, Joseph Weizenbaum created ELIZA, the first chatbot, as an experimental emotional tool. Between 1974 and the early 1990s period, an "AI winter" took place. This term, was first used in 1984 and defined the gap between the AI expectations and the technology’s shortcomings. AI gradually restored its reputation in the late 1990s and early 21st century by exploiting formal mathematical methods and by finding specific solutions to specific problems. By 2000, solutions developed by AI researchers were widely used. Deep learning began to dominate industry benchmarks in 2012 and was adopted throughout the field. Neural-network approaches also advanced natural language processing. In 2013, word2vec introduced efficient methods for learning distributed word representations. Beginning in the mid-2010s, generative AI systems began to rapidly improve. In 2014–2015, sequence-to-sequence and attention mechanisms advanced neural machine translation. In 2014, the introduction of models such as the variational autoencoder (VAE) and generative adversarial network (GAN) enabled effective deep generative modeling of complex data such as images. In 2017, researchers at Google Brain introduced the transformer architecture in a paper titled "Attention Is All You Need". The transformer is based on self-attention, which later became the basis of many large language models. Large language models are trained on vast quantities of data and can be adapted to a wide range tasks. They can discuss a huge number of topics and display general knowledge. Investment in AI grew exponentially after 2020, with venture capital funding for generative AI companies increasing dramatically. Total AI investments rose from $18 billion in 2014 to $119 billion in 2021, with generative AI accounting for approximately 30% of investments by 2023. The commercial AI scene became dominated by American Big Tech companies, whose investments in this area surpassed those from U.S.-based venture capitalists. OpenAI's valuation reached $86 billion by early 2024, while NVIDIA's market capitalization surpassed $3.3 trillion by mid-2024, making it the world's largest company by market capitalization as the demand for AI-capable GPUs surged.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

GPT-3 is a large language model that was released in 2020 by OpenAI and is capable of generating high-quality human-like text. The text-to-image models DALL-E 2 and Midjourney were released in 2022. ChatGPT, an AI chatbot created by OpenAI, launched in November 2022 and became the fastest-growing consumer software application in history, gaining over 100 million users in two months. The success of ChatGPT brought generative AI into the public consciousness, popularizing the use of generative AI for general-purpose text-based tasks.

As time passed, the power of generative AI grew stronger. In 2015, initial popularity began to grow with the release of Google's DeepDream. DeepDream is a generative AI that takes inputs from a previous image and morphs them to produce hallucinogenic images. In January 2021, OpenAI released DALL-E, allowing for image generation through text prompts. This allows users to generate any image with a simple prompt. Soon after, other powerful models followed DALL-E, such as Google's Gemini. The popularity of text-to-video generative AI tools grew exponentially. With the release of models such as OpenAI's Sora in 2024, the use of text-to-video tools became normalized, as people used them for advertisements, which saves on production costs and increases production speed. Generative AI is growing at a rapid rate, outpacing modern-day detection tools. With the common public having access to these tools, it raises concerns about the ethical use of generative AI. There have been multiple occasions where misinformation has been spread over the internet about politics due to a generated or deep-faked video, posing as a security threat.

Generative coding can be used to produce, edit, explain, and debug code. A 2026 study in the journal Management Science found that less experienced developers have higher adoption rates and greater productivity gains.

The use of AI leads to a request and the needs of more energy. To meet this enormous demand, engineering realities show that the U.S. gas production will increase by 10 to 20% by 2030, as most of the additional electricity generation will mostly come from burning natural gas. Then, after 2030, those demands will probably be met by additional solar capacity and nuclear energy Electricity consumed by hardware used for AI has increased demands on power grids, which has led to prolonged use of fossil fuel power plants which would otherwise have been deactivated, since renewable energy sources often lack the continuous baseload power required for 24/7 AI model training and inference. Microsoft, Google, and Amazon have all invested in existing or proposed nuclear power plants to meet these demands. In September 2024, Microsoft signed a deal with Constellation Energy to purchase power from a reactor at Three Mile Island which had been shut down in 2019. The reactor is set to reopen in 2028 to provide power to Microsoft's data centers. The reactor is next to the unit which caused the worst nuclear power accident in US history in 1979.

Some economists have been optimistic about the potential of the current wave of AI to boost productivity and economic growth. Notably, Stanford University economist Erik Brynjolfsson, in a series of articles has argued for an "AI-powered Productivity Boom" and a "Coming Productivity Boom". At the same time, others like Northwestern University economist Robert Gordon remain more pessimistic. Brynjolfsson and Gordon have made a formal bet, registered at long bets, about the rate of productivity growth in the 2020s, to be resolved at the end of the decade. Big Tech companies view the AI boom as both opportunity and threat; Alphabet's Google, for example, realized that ChatGPT could be an innovator's dilemma-like replacement for Google Search. The company merged DeepMind and Google Brain, a rival internal unit, to accelerate its AI research. The market capitalization of Nvidia, whose GPUs are in high demand to train and use generative AI models, rose to over US$3.3 trillion, making it the world's largest company by market capitalization as of 19 June 2024 and became the first company to reach US$4 trillion on 9 July 2025 and subsequently US$5 trillion on 29 October later that same year. In 2023, San Francisco's population increased for the first time in years, with the boom cited as a contributing factor. Machine learning resources, hardware or software can be bought and licensed off-the-shelf or as cloud platform services. This enables wide and publicly available uses, spreading AI skills. Over half of businesses consider AI to be a top organizational priority and to be the most crucial technological advancement in many decades. Across industries, generative AI tools are becoming widely available through the AI boom and are increasingly used in businesses across regions. A main area of use is data analytics. Seen as an incremental change, machine learning improves industry performance. Businesses report AI to be most useful in increased process efficiency, improved decision-making and strengthening of existing services and products. Through adoption, AI has already positively influenced revenue generation in multiple business functions. Businesses have experienced revenue increases of up to 16%, mainly in manufacturing, risk management and research and development. AI and generative AI investments have been increasing with the boom, increasing from $18 billion in 2014 to $119 billion in 2021. Most notably, the share of generative AI investments was around 30% in 2023. Further, generative AI businesses have seen considerable venture capital investments even though regulatory and economic outlooks remain in question. Tech giants capture the bulk of the monetary gains from AI and act as major suppliers to or customers of private users and other businesses. With the introduction of AI, there has been an exponential rise in production for businesses. It's expected that workers could use resources provided by artificial intelligence to boost their productivity. As many small businesses don't use AI, it's believed that if it's adopted by more businesses, the whole work structure could be changed, as many tasks will be automated by AI. The U.S. Census Bureau's Business Trends and Outlook Survey measured AI use at 3–9% of businesses in March 2025, using language that asked whether firms used AI "to produce goods and services". After revising the question to cover "any business function", the Bureau's measured adoption rate increased to 18% in March 2026. According to Bell & Korinek (2023), the economic effect of AI could worsen economic inequality, which could potentially threaten democracy in ways which are separate from threats caused by AI's use in misinformation and propaganda. As of second quarter 2026, Nvidia is the world's largest corporation by capitalization, and semiconductor manufacturers TSMC, Broadcom, and Samsung are among the ten largest.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

Inaccuracy, cybersecurity and intellectual property infringement are considered to be the main risks associated with the boom, although not many actively attempt to mitigate the risk. Large language models have been criticized for reproducing biases inherited from their training data, including discriminatory biases related to ethnicity or gender. As a dual-use technology, AI carries risks of misuse by malicious actors. As AI becomes more sophisticated, it may eventually become cheaper and more efficient than human workers, which could cause technological unemployment and a transition period of economic turmoil. Reaction to the AI boom has been mixed, with some hailing the new possibilities that AI creates, its sophistication and potential for benefiting humanity; while others denounced it for threatening job security and for giving 'uncanny' or flawed responses. Some observers argue an AI arms race or AI Cold War exists between the US and China. The term "AI Cold War" was coined in 2018 by Ian Bremmer and Nicholas Thompson in a Wired magazine article. Kai-Fu Lee, the president of Google China from 2005 to 2009, made a similar claim in his 2018 book AI Superpowers, further arguing "If data is the new oil, then China is the new Saudi Arabia."

Tech companies such as Meta, OpenAI and Nvidia have been sued by artists, writers, journalists, and software developers for using their work to train AI models. Early generative AI chatbots, such as the GPT-1, used the BookCorpus, and books are still the best source of training data for producing high-quality language models. ChatGPT aroused suspicion that its sources included libraries of pirated content after the chatbot produced detailed summaries of every part of Sarah Silverman's The Bedwetter and verbatim excerpts of paywalled content from The New York Times. In protest of the UK government holding consultations on how copyrighted music can legally be used to train AI models, more than a thousand British musicians released an album with no sounds, entitled Is This What We Want?

The ability to generate convincing, personalized messages as well as realistic images may facilitate large-scale misinformation, manipulation, and propaganda. On 19 April 2024, as part of an ongoing feud with fellow rapper Kendrick Lamar, the artist Drake released the diss track "Taylor Made Freestyle", which featured AI-generated vocals imitating the voices of Tupac Shakur and Snoop Dogg. Shakur's estate threatened to sue over the use of Shakur's likeness, saying that it constituted a violation of Shakur's personality rights. On 20 May 2024, following the release of a demo of updates to OpenAI's ChatGPT Voice Mode feature a week earlier, actor Scarlett Johansson issued a statement in relation to the "Sky" voice shown in the demo, accusing OpenAI of producing it to be very similar to her own, and her portrayal of the artificial intelligence voice assistant Samantha in the film Her (2013), despite Johansson refusing an earlier offer from the company to provide her voice for the system. The agent of the unnamed voice actress who voiced Sky stated that she had recorded her lines in her natural speaking voice and that OpenAI had not mentioned the movie Her nor Johansson. Several incidents involving the sharing of non-consensual deepfake pornography have occurred. In late January 2024, deepfake images of American musician Taylor Swift proliferated. Several experts have warned that deepfake pornography is more quickly created and disseminated, due to the relative ease of using the technology. Canada introduced federal legislation targeting sharing of non-consensual sexually explicit AI-generated photos; most provinces already had such laws. In the United States, the DEFIANCE Act was introduced in March 2024.

A large amount of electricity is needed to power generative AI products, making it more difficult for companies to achieve net zero emissions. From 2019 to 2024, Google's greenhouse gas emissions increased by nearly 50%, partly as a result of increased energy consumption by AI data centres.

Industry leaders and others have signed the Statement on AI Risk, arguing that humanity might irreversibly lose control over a sufficiently advanced artificial general intelligence (AGI).

5. Training Protocol, Convergence Behavior & Experimental Evaluation

Much of the AI boom has been funded by loans and venture capital, but many commercial AI services remain of questionable practical utility or quality for business. Despite more than $60 billion in corporate investment in AI in 2025, 95% of business AI projects are unprofitable, according to research from MIT. Producers of generative AI, such as OpenAI, also themselves currently have costs greatly exceeding their revenue. As other major tech companies such as Nvidia are both heavily invested into AI and dependent on the AI ecosystem and its hardware demands for their own ongoing growth, this has raised speculation of a wider economic bubble in the tech industry, particularly if future demand falls short of the current levels of AI investment.

AI bubble AI datacenter AI effect AI slop AI winter, a period of reduced funding and interest in artificial intelligence research History of artificial intelligence History of artificial neural networks Hype cycle List of artificial intelligence projects List of artificial intelligence journals Lists of open-source artificial intelligence software Progress in artificial intelligence Regulation of artificial intelligence Technological singularity Timeline of artificial intelligence risks in global finance

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .