How to Reduce Energy Consumption with AI: Practical Prompts

How to Reduce Building Energy Consumption with AI: A Quick Guide

Commercial buildings account for over 40% of global energy consumption each year. Fortunately, large language models (LLMs) and prompt engineering tools now offer practical ways to cut these costs while maintaining comfort. In this article, you’ll find actionable prompts, code examples, and real-world use cases you can start applying today.

Why AI is Essential for Green Tech Today

• REAL-TIME PREDICTION:LLMs analyze sensor data on consumption, outdoor temperature, and occupant schedules to forecast demand spikes.
• ADAPTIVE AUTOMATION:Agents can dynamically adjust lighting, HVAC, and IT equipment settings without human intervention.
• TRANSPARENCY AND COMPLIANCE:

What Prompt to Use for an Intelligent Energy Management System

Below is a ready-to-use prompt that you can copy into a chat interface and adapt to your building or facility.

Design a Python script that uses the OpenAI API to retrieve sensor consumption data (kWh, temperature, humidity) from a timeseries database (e.g., InfluxDB) for the last 24 hours. Then generate a prompt for an LLM that:
1. Identifies peak periods (>10% above hourly average).
2. Suggests HVAC adjustments (temperature setting, fan speed) to reduce load during peaks.
3. Returns a JSON schedule with timestamps of adjustments and new setpoints.

Include an example JSON response and explain how to verify that the changes are safe (e.g., keeping indoor temperature between 20-24°C).

The prompt is deliberately specific: it defines the input data, the task, the output format, and safety rules. This increases the likelihood of receiving a functional script in a single response.

Code Example: Predictive Control with LLM

Below is a minimal workflow you can run in a notebook. It uses OpenAI API calls and an InfluxDB client.

import os
import json
import requests
from influxdb import InfluxDBClient

# Configure clients
influx = InfluxDBClient(host='your-influx-host', port=8086, token='your-token', org='your-org')
openai_api_key = os.getenv('OPENAI_API_KEY')

# Retrieve sensor data for the last 24 hours
query = '''
from(bucket="sensors")
  |> range(start: -24h)
  |> filter(fn: (r) => r._measurement == "power" or r._measurement == "temperature")
  |> aggregateWindow(every: 1h, fn: mean)
'''
result = influx.query(query)
points = list(result.get_points())

# Prepare context for the LLM
context = json.dumps(points, indent=2)

# Call OpenAI API
headers = {'Authorization': f'Bearer {openai_api_key}', 'Content-Type': 'application/json'}
payload = {
    "model": "gpt-4o",
    "messages": [
        {"role": "system", "content": "You are an energy engineer specializing in LLM-based optimization."},
        {"role": "user", "content": f"Analyze this consumption and temperature data: {context}. Identify consumption peaks and suggest HVAC adjustments. Return a JSON with recommended HVAC setpoints."}
    ],
    "temperature": 0.1
}

resp = requests.post('https://api.openai.com/v1/chat/completions', headers=headers, json=payload)
llm_response = resp.json()['choices'][0]['message']['content']
# Extract JSON (LLMs often wrap JSON in a code block)
import re
json_match = re.search(r'\{[\s\S]*\}', llm_response)
if json_match:
    schedule = json.loads(json_match.group())
    print(json.dumps(schedule, indent=2))
else:
    print("No valid JSON returned")

How to Integrate the Latest AI Technologies into Your Green Tech Stack

• Voice Commands with Soniox STT v6:Use Soniox’s new Echo Filtering feature to issue voice commands like "Increase temperature by 2°C" without interference. Integrate the transcribed text with the script above for voice-based energy control for building managers.
• Model Integrity Guarantee with FlashProve:
• Faster Deployment with Pando:Upload the script and change logs so that Pando handles branches and packfiles efficiently. Result: LLM model updates for energy optimization are deployed with sub-second latency, keeping control loops in real time.

Real-World Use Cases (2026)

Smart Office Building:A 30-story skyscraper in the United States implemented the LLM script to reduce peak energy consumption by 12% in three months. They integrated Soniox voice commands to allow facility managers to quickly adjust setpoints during emergency meetings.

Factory 4.0:A European semiconductor manufacturer used LLM predictive control to dynamically adjust machine cooling systems. By combining FlashProve with their models, they achieved regulatory compliance and a 9% reduction in energy costs.

Concrete Actions You Can Take Today

1. Create a basic prompt (see above) and test it with a small set of historical data.
2. Try a voice command using Soniox STT v6 to turn lighting on/off; measure latency.
3. Sign your LLM model with FlashProve before any production deployment.
4. Deploy the code with Pando to automate CI/CD testing and reduce update latency.

Conclusion: AI is the New Lever for Energy Efficiency