Bhubaneswar: In a proud moment for Odisha and the global tech community, Odia software engineer and entrepreneur Sam Pasupalak, co-founder and CEO of Skyfall AI, has developed an innovative AI model named SKY-001.
The model is designed specifically for computer-aided design (CAD) and building information modelling (BIM), generating structured engineering designs rather than generic outputs.
Developed over 18 months, the system aims to resolve common limitations in generative design by breaking complex projects down into progressive stages, mapping individual components, assemblies, and construction steps before execution rather than generating full designs at once.
According to benchmarks shared by Pasupalak on X (formerly Twitter), SKY-001 dramatically outperforms current frontier systems, including GPT-6 Astra, on complex spatial and structural engineering tasks. While traditional large language and multi-modal models such as ChatGPT, Gemini, or Claude, which rely heavily on text retrieval from websites like Reddit, Wikipedia, and Twitter, often struggle with the rigid geometric, dimensional, and long-horizon planning constraints required in manufacturing and architecture, SKY-001 approaches the challenge as a sequential visual-spatial problem.
Explaining the core philosophy behind SKY-001, he highlighted its unique ability to simulate human-like foresight and planning. “At Skyfall, our mission is to build completely autonomous software, which can imagine, plan and execute like humans. As human beings, we try to dream and imagine hundreds or thousands of steps ahead before taking an action in the real world… We built that human ability to plan in the long term into a world model. LLMs can never get us there since they fail at long horizon planning because they are constrained by static web text and can plan only ten to a hundred steps at best,” he said.
While unveiling Sky 001, Pasupalak described it as the first foundational world model for engineering, which enables users to visualise future steps and design anything they imagine in engineering and software fields through an entirely new technological approach.
Building a cutting-edge AI model typically requires massive computing power, with major tech models trained on hundreds of thousands of GPUs. However, Pasupalak’s team accomplished this breakthrough under significant hardware constraints. “Obviously, we didn’t have those many GPUs, so that was a very constraining factor… We had to focus a lot on the efficiency aspect, and on speed and efficiency,” he shared.
Developed over 18 intense months by a lean team of 25 members spread across India, San Francisco, and Toronto, SKY-001 delivers high-speed, cost-effective performance that rivals leading global AI systems at a fraction of the cost. “Sky 001’s world model is at least 100X faster, 150 X cheaper than GPT-Astra and all the other frontier models out there while having much higher precision than all of them,” he said.
According to the CAD cost comparison shared by the founder, running a sample on SKY-001 costs just $0.005, compared to $0.75 on GPT-6 Astra. This makes GPT-6 Astra approximately 150 times more expensive than SKY-001 for this task.
He elaborated further using the example of a high-performance sports car. “Sky 001 allows you to dream what a sports car would look like and saves thousands of hours using CAD software. First, you explore multiple designs of the car engine in parallel, all at once. The car engine is now broken down part by part, hierarchically into its smallest component with the highest precision. The 3D construction world model assembles each part back into the 3D space so that you can now visualise using CAD software like Autodesk Fusion. We tried our world model with building Architecture software and it worked just like magic,” he said.
Looking ahead, Pasupalak stated that while the startup is currently focused on foundational steps in CAD and BIM, the underlying technology holds immense potential for expansion into multiple verticals—including PCB design, aerospace engineering, molecular design, protein synthesis, and drug discovery.
A native of Bhubaneswar, he holds a bachelor’s degree in Mathematics and Computer Science from the University of San Francisco (2014) and a master’s degree in Political Science from the London School of Economics. His journey in artificial intelligence began over 15 years ago, during the ‘AI winter’, working alongside Turing Award winners Prof Yoshua Bengio and Prof Richard Sutton. As early pioneers in deep learning, they built Maluuba, an original AI personal assistant that surpassed Siri in capability. Microsoft acquired Maluuba just a year before the release of GPT-1. Looking back, he admits to a touch of regret over the sale, knowing Maluuba had the potential to become what OpenAI is today. But for him, the past is a prelude: the time has come to build the next trillion-dollar company.
He is the son Dr S N Pasupalak, former Vice-Chancellor of Odisha University of Agriculture and Technology (OUAT), Bhubaneswar.
Offering a message to young aspirants in India and Odisha, Pasupalak emphasised that modern technology is fully democratised: “It doesn’t matter if you are in India, Odisha, or Japan anymore… You just have to know what problem you want to solve, figure it out, and push yourself hard to achieve the impossible.”
