Buildings 2.0
In this episode of Buildings 2.0, Jose speaks with Mayur Mistry, Co-founder, 3DGuru.ai and 3DGuru MVP Labs. Through his organizations, Mayur is democratizing AI tools for architecture and construction firms. He shares eye-opening insights about how the technology landscape has evolved, from needing 5,000 images to train an AI model in 2019, to now requiring just 5 images and a few dollars. Through his consulting work, he's also tackling a crucial challenge: helping medium-sized firms innovate without the hefty budgets typically needed for internal tool development.
Mayur explores practical frameworks for AI adoption, the importance of specialized data in beating bigger competitors, and why firms need to think strategically about their data today to stay ahead tomorrow. Mayur also shares valuable insights about different AI models (Claude vs. OpenAI vs. Llama) and how firms can choose the right tools for their specific needs.
Topics discussed:
- How AI tools have become radically accessible going from requiring 5,000 images and specialized knowledge in 2019 to now needing just 5 images and the cost of a coffee.
- Strategic approaches for firms to adopt AI: starting with pilot projects, focusing on immediate ROI, and leveraging no-code/low-code solutions for quick implementation.
- Why firms need a robust data strategy today, including organizing unstructured data, leveraging specialized knowledge, and preparing for future AI implementation.
- AI model selection, such as when to use Claude for coding tasks, OpenAI for general applications, and Llama for data-sensitive operations within firms.
- Bootstrapping lessons learned on why starting with service-based consulting helped validate market needs and generate immediate revenue while building the product.
- Breaking down AI agents, including how natural language can now orchestrate complex workflows, from real estate prospecting to automated design tasks in architecture.
Intro Quote:
“Earlier in 2018 and '19, if I had to train an AI model, let's say using generative adversarial network I had to spend at least a thousand to five thousand images of the firm, need a good GPU, and then we'll get a model. And it requires specialized knowledge. Now I can train a model using five images and it will cost me less than a Starbucks coffee, a few dollars.” 7:27-7:59
Get in touch with Mayur Mistry:
Get in touch with your host, Jose Luis Cruz:
Listen to more episodes:
