Community
28 February 2025
Multimodal AI Agents: two-day Hackathon with Weights & Biases, Google Deepmind
Hyperpocket, VESSL AI’s new open-source project, recently hosted a dynamic hackathon in collaboration with Weights & Biases (W&B) and Google DeepMind.
The event received additional support from various agent industry leaders. Over two days, participants showcased creative applications of AI agent —featuring multimodal abilities and tool integrations—while industry experts shared valuable insights on the evolving role of AI agents.
Hyperpocket’s Vision: Why Tool-Calling Matters
During the hackathon, **Lidia Jaewon Yoo, Product Manager at Hyperpocket**, delivered a compelling presentation on the importance of integrating external functions (tool-calling) into AI agents. Through a practical demo, she demonstrated how Hyperpocket seamlessly integrates external functionalities into AI agents.
Key Insights
Seamless Tool Integration
- Hyperpocket makes it easy to develop, share, and integrate these tools, allowing teams to expand their agents’ capabilities with minimal effort.
- Tools function as the action layer for AI agents, enabling them to connect with external systems, APIs, and logic.
Simplified Authentication
- Hyperpocket supports various authentication methods, such as token-based authentication and OAuth2, and offers built-in mechanisms to streamline the process.
- When calling external tools, secure authentication is crucial to regulate access.
Highlight: On-Demand Tool Development in Just 1 Hour
One of the most exciting highlights was the real-time tool development at Hyperpocket’s booth. Thanks to Hyperpocket’s user-friendly design and large tool library, the team could meet participants’ requests quickly—some tools were already built, and new ones were created and delivered within an hour.
New Tools from the Hackathon
- Zinc – Zinc is a shopping automation tool that integrates Walmart or Amazon ordering APIs.
- Semantic Scholar – Semantic Scholar is a research tool that helps users locate and analyze scientific publications, including authors, papers, citations, and venues.
- AgentQL – AgentQL is a data extraction tool that retrieves structured information from web pages, PDFs, and images through HTTP queries.
"This rapid development demonstrates Hyperpocket’s flexibility, allowing developers to quickly create and integrate the necessary tools, keeping their focus on building AI agents themselves”
Highlight: Best Hyperpocket Use Case Award – "People Coming Over”
The Best Use Case of Hyperpocket Award went to Team "People Coming Over”, which captivated participants and judges alike.
- Project Overview
- Their AI agent analyzes a user’s room from an uploaded image, then provides tailored recommendations to make the space cozier and more welcoming.
- Role of Hyperpocket
- They needed a tool to pull product data for the AI-driven suggestions.
- Collaborating with the Hyperpocket team, they integrated the brand-new Zinc tool into their agent.
- Why They Won
- They not only built a compelling agent experience but also offered thoughtful feedback, actively helping improve the platform by identifying bugs and usability issues.
Key Takeaways
1. Preferred Features for AI Agent Developers
- Many participants prioritized selecting an LLM known for strong reasoning and decision-making.
- After picking an LLM, they explored various tools to boost automation and interactivity.
- If a desired tool wasn’t readily available, they built a custom tool to meet specific project needs.
2. Customizing a Tool for an Agent
- The simplest approach is to create a function that includes the necessary logic, then make it function-calling tool using decorator of it.
- Furthermore for scalability and reusability, consistent design patterns are crucial.
- Several participants built and integrated their own tools, reflecting the adaptability of Hyperpocket’s architecture.
Final Thoughts
The Multimodal Agent Hackathon was a great success, showcasing Hyperpocket as a versatile and developer-friendly platform for AI agent creation. The event encouraged collaboration and highlighted how swiftly custom tools can be built, integrated, and scaled.
Hyperpocket will continue to introduce seamless integrations and robust tool-calling features, offering even more options for those building AI agents. With an active community and a drive for rapid innovation, Hyperpocket is poised to become a key player in the future of agent-centric AI applications.
Stay tuned as Hyperpocket grows, making AI agent building easier!
Sanghyuk
ML Engineer
Lucas
Software Engineer
Kelly
Head of Growth
Simon
Business Development Manager
Try VESSL today
Build, train, and deploy models faster at scale with fully managed infrastructure, tools, and workflows.