Smart Model Ensembles for Smarter Construction

Deploying machine learning models in the construction industry is challenging and requires a lot of manual work. Custom code should only be built when necessary and should rely on open source. Machine learning is far from automated and still requires manual work. The construction industry is averse to innovative solutions and change.

• The construction industry is averse to innovative solutions and change
• Custom code should only be built when necessary and should rely on open source
• Machine learning is far from automated and still requires manual work
• Deploying machine learning models in the construction industry is challenging and requires a lot of manual work

The speaker was shocked to find that people in the construction industry still emailed each other PDF files instead of using a digitally first format. One company the speaker talked to still had a CRM system running on DOS and was quite happy with it. This illustrates the resistance to change in the construction industry.

AI is spurring a new wave of companies in construction tech that are transforming this $10 trillion-a-year industry. There’s substantial margin risk in construction when projects aren’t well planned, and this eats into economic growth, because construction represents 14% of the world’s GDP. The beginning of a project, known as “take-off,” is critical. Oleksandr Paraska, CTO at Togal.ai, discusses how some of the best machine learning and computer science experts are improving efficiency in construction. Oleksandr talks about his journey in guiding his startup by integrating domain experts (i.e. architects) and engineering skill sets when building models. He also explains how the nuances in floorplan (architecture) diagrams require object detection, classification, and segmentation to make sense of. He also shares the challenges of scaling up processing volumes, in spite of the limits of GPUs, inference times, infrastructure cost and MLOps resourcing.