Insights from Modal's Hackathon: Projects Unveiling New Use Cases and Enhancements

During a recent offsite in Ericeira, Portugal, Modal hosted an internal hackathon that brought together 30 team members divided into 13 groups. Their mission was to develop projects utilizing Modal, with a focus on applications such as AI agents, real-time translation, and music platforms. The results not only highlighted the creativity of the teams but also unveiled new ways to integrate Modal into production-oriented environments. This type of hands-on event fosters not only technical innovation but also strengthens team cohesion, allowing members to collaborate and iterate on ideas rapidly.

Browserman: A Multi-Modal AI Agent

The Browserman team developed a multi-modal AI agent capable of navigating the internet using visual cues to accomplish user-defined tasks. For example, when tasked with reordering a favorite meal from Domino's, Browserman autonomously locates and interacts with the Domino's website, navigating through ads and options to fulfill the request. This not only showcases user-directed capability but also emphasizes the potential for practical applications of AI in everyday scenarios, where automation can save consumers time.

This application is built using Llama-3.2-90B-Vision-Instruct-FP8, vLLM, and Playwright. It utilizes Modal’s distributed Queues to manage requests efficiently. Interestingly, this innovative approach emerged just before a similar announcement from Anthropic, marking a competitive edge for Browserman. The timing indicates a strategic awareness of the competitive landscape, which could give Modal an advantage in future negotiations or partnerships.

Glodal: Visualizing Modal's Global Reach

The Glodal team focused on creating an interactive visualization tool to showcase how client requests travel across Modal's global infrastructure. By illustrating the journey from a user's initial request to the Modal control plane in the U.S., and then to the worker nodes across Europe and Asia, Glodal underscores the importance of efficient data routing. This sort of visualization is not only informative for users but also critical for engineers seeking optimization opportunities within the network.

Central to this project is the demonstration of a distributed control plane, which could potentially reduce latency by a factor of 100. The existing architecture requires transatlantic communication for requests originating from Europe, leading to latency times around 268ms. In contrast, the new distributed control setup can keep communications within the continent, reducing this to about 3ms. This latitudinal shift could greatly enhance user experiences, particularly for applications requiring real-time data processing:

• A client in eu-west sends a request to the us-east control plane.
• The control plane then redundantly triggers worker nodes in eu-west, incurring additional latency.

Glodal’s findings could lead to significant performance enhancements across Modal’s services, with ongoing efforts to integrate a distributed control plane into the platform. If you're working in this space, the implications are clear: faster responses and improved performance could keep Modal competitive in an already crowded field.

Waluigi: Integrating Modal with Workflow Orchestration

Originators of the Luigi workflow orchestrator, Erik Bernhardsson and Elias Frieder, aimed to integrate Modal as a worker node within Luigi's framework through their project, Waluigi. This combination enhances batch processing by maximizing Modal's computational capabilities. The visualization below shows how a Luigi Directed Acyclic Graph (DAG) operates with Modal tasks running simultaneously:

This integration fosters several advantages:

• Auto-scaling to maximize parallel task execution offers better resource management.
• Customization of each task’s hardware requirements, including GPU support, allows for task optimization based on specific needs.
• Orchestration of task dependencies paves the way for efficient workload distribution and initiatives for partial retries during executions, reducing the impact of task failures.

pytest-modal: Enhancing Testing Efficiency

The pytest-modal team focused on creating a plugin that enables parallel execution of test suites on Modal. By running individual tests in separate containers, they utilized Modal's capacity to manage bursty workloads effectively. This approach could not only enhance the efficiency of testing protocols but also ensure rapid deployment cycles, a critical factor in software development:

Implications and Outlook

The presentations at the hackathon showcased the ingenuity and resourcefulness of employees, demonstrating what can be achieved when creative minds are given the right tools and environment. These projects highlight a market-ready, extensible framework that can be adapted to a variety of applications, from AI agents to testing and orchestration.

However, the impact of these innovations won't only be limited to Modal's internal processes; they could redefine how developers interact with cloud infrastructures. The shift towards personalizing services and reducing latency can directly improve user satisfaction—a factor that cannot be ignored in today’s competitive tech environment.

What this means for you as a developer is clear: the promise of faster, more efficient tools is becoming a reality. If you're working with Modal's platform, now's the time to explore how these new tools can streamline your workflows. In an industry that demands quick turnarounds, the ability to quickly adapt and implement such technologies will be invaluable.

The Modal team over time