Intel AI Startup Olympics: Hackathon to Accelerate Innovation

Author: Eduardo Alvarez

Intel recently hosted the AI Startup Olympics, a 3-day event focused on implementing Intel-accelerated software and hardware in industry-leading solution architectures. The Hackathon challenge consisted of building a machine learning application according to specifications defined in the “Solution Blueprint.” The blueprint provides enough flexibility for teams to bring their data and code to the hackathon. Teams were judged on their ability to build according to these specifications, solution innovation, and workload optimization.

This event was hosted through the Intel® oneAPI for Startups program, which is focused on startups specializing in deep technical solutions with the resources to match.

All participants had free access to the latest 4th Gen Intel® Xeon® processors in Bare Metal Instances, dual-socket 56-core machines with two threads per core, on Intel® Developer Cloud. This gave startups an opportunity to test their own data and code on the latest Intel hardware and to be able to take advantage of accelerations like Intel® Advanced Matrix Extensions (Intel AMX).

What was this hackathon all about?

The hackathon focused on building an AI application using a microservice architecture design pattern. The competition aimed to give startups an opportunity to showcase their skills, and to incorporate components of the Intel® AI Analytics Toolkit and hardware optimizations into their existing product code-bases. The goal of the event was to promote AI innovation and to encourage startups to leverage the latest technologies to create cutting-edge AI applications. The competition provided a forum for participants to collaborate, learn from one another, and showcase their creativity and technical abilities.

During the hackathon, the participants were required to adhere to a set of technical requirements. These requirements included splitting the application into separate processes with at least two microservices such as data wrangling/processing, training and validation, and inference. Additionally, each microservice had to be deployed as a containerized application using Docker* or other popular container management tools. Furthermore, at least one of the two microservices had to leverage an Intel deep learning accelerated extension such as AMX. Each microservice had to contain an API end-point that could receive HTTP requests in JSON format. Lastly, the only compute resources and storage available for the solution were the 4th Gen Xeon Bare Metal Intel Developer Cloud Instance.

Teams had the opportunity to try out some cool side quests for bonus points. These quests included incorporating a working database layer, adding event messaging, using OpenMP* topology optimizations, leveraging an Intel-accelerated extension in both main microservices, building a model observability or performance tracking microservice, and including a performance benchmark report in the final presentation. Optional side quests allowed the teams to explore different areas of innovation and showcase their skills in various domains, making the hackathon a unique and exciting experience for all participants.

Learn more about the innovative startups that participated

The hackathon event was exclusively designed for members of the oneAPI Startup Program, hosting four members at this invite-only event: Cerebra, Selecton Technologies, Sefi AI, and Beewant. Following the hackathon, we had the opportunity to sit down with three of the participating startups to learn more about their companies, their experiences during the event, and their vision for collaborating with Intel’s oneAPI for Startups program. Here is a summary of each:

Selecton is a generative AI startup specializing in natural language processing and large language models that help customers make better decisions for a more personalized buying experience. They use reinforcement learning to enrich the concept and resolve ambiguity. During the AI Startup Olympics, the Selecton team learned about microservices and architecture types, and used Intel tools to optimize their models. The collaboration with Intel helped them build a reliable and scalable infrastructure in a cost- and performance-efficient way. The team is excited about future opportunities to network with Intel Capital, co- publish their work, and use Intel DevCloud for their research and production.

Beewant offers a distinct approach to processing images, videos, and text that enables organizations to explore their vast amounts of unstructured data. Their technology is multimodal and allows for intuitive natural language searching of concepts and information. During the AI Startup Olympics, Beewant had an opportunity to showcase their search solution and refine their product. The event gave the team members valuable experience in utilizing their expertise in deep learning and micro-services to work collaboratively to deliver tangible results. The AI Startup Olympics helped the team stay abreast of industry trends, improve their communication skills, and take their MLOps to a new level.

Cerebra develops medical software that uses machine learning algorithms to quickly diagnose strokes and to prevent neurological deficiency and brain cell death. The software produces an analysis within three minutes and helps triage acute stroke patients, leading to timely and accurate diagnosis and improved patient outcomes. During the hackathon, the machine learning department received valuable support and guidance on Intel's optimization tools, leading to the discovery of cost-saving techniques such as cutting GPU usage. Working with Intel has allowed Cerebra to improve their deep learning models, and the hackathon provided them with a unique experience to collaborate with other AI startups.

The AI Startup Olympics enabled these AI startups to showcase their technology solutions, refine their products, and collaborate with Intel to improve their deep learning models. They gained insights on solution architecture, Intel products, modern deployment practices, and developing consistent infrastructure around their models. Working with established technology companies such as Intel provided these startups with resources, expertise, and technology solutions to develop and scale their offerings more quickly. Access to the latest hardware and software solutions, such as Intel 4th Gen Xeon processors and AI software tools and frameworks, can help AI startups stay competitive in the fast-evolving field of AI.

If you are interested in learning more details about the event and the challenge, you can visit the event GitHub* repository. And here was the agenda:

What’s next?

Our team plans to continue to create opportunities to collaborate in fun developer-first events like the AI Startup Hackathon. We will likely be kicking off a second startup-focused hackathon in the second quarter of the year. If you would like to get access to cloud resources to start building with the latest hardware and software available, you can create an Intel Developer Cloud account at cloud.intel.com. If you would like to learn more about joining a startup community like this, I encourage you to check out A Guide to AI Partner and Accelerator Program for Startups.