Two Katz School AI students took first prize at the 2023 UC Berkeley AI Summit Generative AI Hackathon for solving a complex business problem using the latest in generative AI in under two hours.

Niranjan Kumar Kishore and Tharun Prabhakar won in the category of “College Student, Data Scientists or Comp Sci,” with Prabhakar completing his generative AI application just two minutes behind Kishore in the contest. Both were awarded $1,000.

Their winning submissions were an end-to-end automated generative AI application implemented in a secure Google Cloud that used large language models to answer a set of questions on the impact, opportunity and challenges of generative AI, what every CEO should know about generative AI, and generative AI in healthcare.

Kishore and Prabhakar competed against over 100 students and working professionals representing the companies Broadcom, Ciena, Cisco, Electronic Arts, Johnson & Johnson, Meta, Optum and Wipro, as well as Arizona State University, Babson College, University of California Davis and University of California Berkeley. The competition was hosted by Aible and judged by leaders in technology and AI from UC Berkeley, Babson College, University of Colorado and Google.

Generative AI refers to a class of AI models and algorithms that can generate new, original content, such as text, images or other types of data. Unlike traditional AI systems that may rely on rule-based programming or supervised learning from labeled datasets, generative AI is often associated with unsupervised learning and creativity.

“GenAI apps with prompt augmentation and few-shot learning functionality are applications that incorporate advanced techniques to enhance the capabilities and performance of the generative model,” said Kishore, who has a background in biomedical engineering. “These techniques allow users to guide the model through enhanced prompts and by providing it with a limited set of examples, these apps can produce more tailored and contextually relevant outputs.”

One of the key components of generative AI is the use of models, which are trained to understand patterns and structures within data and then generate new data that share similar characteristics. These models are often based on neural networks, particularly generative models like generative adversarial networks (GANs) or variational autoencoders (VAEs).

“We utilized state-of-the-art tools and models for language understanding and generation, enhancing the sophistication and performance of the generative AI solutions,” said Prabhakar. “This experience provided a fantastic opportunity to explore the ever-evolving world of artificial intelligence and its potential in creative problem-solving.”