The Research Behind VERTIQ: How We Automated 82% of LEED Certification

The Problem: Why LEED v5 Certification Still Runs on Spreadsheets

If you have ever managed a LEED certification project, you know the drill. Weeks of cross-referencing floor plans against credit requirements. Manual energy modeling that takes 2-3 weeks just to set up. Dozens of spreadsheets tracking prerequisites and credits across eight categories. And every revision cycle starts the pain all over again.

The LEED certification process has not fundamentally changed in over a decade. Despite advances in AI and automation across nearly every industry, green building consultants still spend 60-70% of their project time on documentation rather than the strategic decisions that actually improve building performance.

We set out to change that. Not with another generic AI tool or a simple template generator, but with a research-backed platform that actually understands construction documents, energy systems, and LEED requirements at a technical level.

The Research: Published, Measurable, Reproducible

VERTIQ is built on research available as a preprint on arXiv [1]: An Integrated Platform for LEED Certification Automation Using Computer Vision and LLM-RAG. The paper, which validated the architecture against LEED v4, presents a modular system that integrates Computer Vision, automated energy simulation, and domain-specific AI to automate LEED credit analysis end-to-end. VERTIQ has since extended this foundation to fully support LEED v5.

This is not a theoretical concept. The system was tested against real construction documents, and every metric was measured against professional benchmarks. Here is what the research demonstrated:

These are not marketing numbers. They come from controlled tests against a 6-story, 75,000 square foot office building with full mechanical, electrical, and plumbing documentation.

How It Works: Three Core Technologies

1. Computer Vision for Construction Documents

Generic AI tools like ChatGPT or Google Gemini struggle with construction drawings. They cannot reliably detect room boundaries, parse hatching patterns, or extract data from mechanical schedules. The accuracy is simply too low for professional use.

VERTIQ uses a purpose-built document processing pipeline. Floor plans are rendered at 600 DPI, then processed through Canny edge detection, connected component analysis, and specialized OCR. The system removes grid lines, hatching patterns, and text overlays that confuse general- purpose vision models, achieving 94% accuracy on text extraction from real construction documents.

This means you can upload your drawings and get accurate room areas, window-to-wall ratios, and mechanical equipment specifications extracted automatically, without manual data entry.

2. Automated Energy Simulation

Traditional energy modeling is the biggest bottleneck in LEED certification. Developing an EnergyPlus baseline model from scratch typically takes 2-3 weeks of a senior engineer's time. That is 2-3 weeks before you even start analyzing EA credits.

VERTIQ automates the IDF (Input Data File) generation process using a framework of specialized extractors: geometric extractors for building envelope data, system extractors for HVAC equipment, and schedule extractors for operational profiles. The system generates a complete EnergyPlus model, runs the simulation, and compares results against ASHRAE 90.1-2022 baselines.

In testing, this reduced energy modeling from 2-3 weeks to under 4 hours, with prediction accuracy within 5% of manually created expert models.

3. Domain-Specific LLM-RAG

If you have tried asking ChatGPT about LEED requirements, you have probably noticed the hallucination problem. General LLMs often cite outdated versions, confuse prerequisites with credits, or simply fabricate point thresholds.

VERTIQ solves this with a Retrieval-Augmented Generation (RAG) system specifically designed for LEED. Instead of chunking documents by sentences or paragraphs (which mixes credit requirements together), the system uses credit-unit chunking: each chunk corresponds to a single LEED credit with its complete requirements, point thresholds, and documentation needs. Combined with vector search, metadata tagging, and post-processing verification, this ensures every generated recommendation is traceable back to the actual LEED reference guide.

From Research to Product

The paper demonstrated the concept. Turning it into a production SaaS platform required significant engineering beyond the research:

• Desktop to cloud: The research prototype ran locally with a PySide6 desktop interface. VERTIQ runs on Google Cloud with a React frontend and FastAPI backend, accessible from any browser.
• Local AI to cloud AI: The prototype used locally hosted models (Gemma 3). VERTIQ uses Vertex AI (Gemini) for reliable, scalable inference with no cold-start delays.
• Single version to multi-version: The research validated the architecture against LEED v4. VERTIQ extended this foundation to LEED v5 BD+C and ID+C, incorporating the v5 decarbonization requirements and updated Platinum criteria.
• Batch to real-time: The research ran analysis in batch mode. VERTIQ provides real-time credit scoring as you update project information, with live progress tracking.

The core modular architecture and its analysis engines remain the same. The infrastructure evolved to meet production requirements: reliability, scale, and user experience.

What This Means for Your Practice

If you are a LEED consultant, the practical implications are straightforward:

• Upload drawings, get credit analysis in hours. Not days. Not weeks. The document processing pipeline extracts what it needs; the energy simulation runs automatically; the LLM-RAG system maps everything to LEED credits.
• Spend time on strategy, not spreadsheets. With up to 82% of credits analyzed automatically, you can focus on the rest that require professional judgment: Innovation credits, design trade-offs, and client-specific optimization.
• Trust the results. Every recommendation links back to published research and the actual LEED reference guide. No black boxes, no hallucinated requirements.

Your first project is free. No credit card required, no feature limitations. See how the research performs on your actual documents.

References

[1] Lee, J. (2025). An Integrated Platform for LEED Certification Automation Using Computer Vision and LLM-RAG. arXiv:2506.00888. https://arxiv.org/abs/2506.00888