GUY | Devpost

Inspiration:

Our inspiration came from research and interviews where we discovered a growing problem: people increasingly rely on AI, often trusting its answers without questioning them. While AI is powerful, it can struggle with highly specialized topics. When that happens, models may generate incomplete, inaccurate, or even fabricated information, which can negatively affect the work people are developing.

We also observed that many users try to solve this by using specialized AI agents. However, these agents usually focus on only one domain. When projects require multiple areas of expertise, the workflow becomes fragmented and incomplete.

This challenge inspired us to rethink how AI could better support complex projects. Instead of relying on a single general model or isolated specialized agents, we envisioned an AI system that could coordinate multiple experts just like a real project manager guiding a team.

What it does:

GUY is an AI agent that acts as a project manager, coordinating multiple specialized AI agents across different domains. Instead of relying on a single model, GUY identifies what the user needs for a specific project and calls the appropriate expert agents to complete each task.

The system focuses on tasks that AI can realistically perform such as planning, research, recommendations, and digital outputs.

Our current prototype focuses on travel planning. GUY coordinates specialized agents to help users design their trips and access the information they need. For example, if a user wants to compare flight prices and discover places to visit, GUY activates the flight and destination agents. If the user only wants hotel recommendations, GUY calls the hotel agent.

Each agent has its own name and role, creating a more intuitive and human centered experience where users feel like they are interacting with a coordinated team rather than a single system.

How we built it

We built ProjectIntelligence as a multi-agent AI travel planning system by converting it from Google Gemini to Amazon Nova Lite. The architecture consists of a React 19 + TypeScript frontend with a Nova-powered orchestrator that acts as an AI project manager, gathering requirements through conversational chat, generating structured project briefs, and coordinating three specialized AI agents

Challenges we ran into:

One of our first challenges was figuring out how to make different AI agents coordinate effectively. At the beginning, we struggled to design a system where one AI could orchestrate and manage the work of other specialized agents without creating confusion or redundant outputs.

Another challenge was designing a more human and comfortable interaction for users. We wanted the experience to feel natural and interactive rather than just a simple text exchange, which required additional work in designing the interface and interaction flow.

We also faced difficulties in helping the project manager agent correctly understand what the user actually needed. Early versions of the system sometimes failed to ask the right questions or clarify missing information, which made it harder to select and coordinate the appropriate specialized agents.

Accomplishments that we're proud of

One accomplishment we are especially proud of is how much we were able to achieve while learning along the way. Although we had programming experience, building a coordinated multi-agent AI system was something completely new to us. What initially seemed very complex gradually became a working prototype through research, experimentation, and persistence.

We are also proud of successfully developing multiple specialized agents, including the project manager agent (GUY), a travel agent(Xime), a restaurant agent(Remy), and a lodging agent(Maximo), and enabling GUY to coordinate them to support the user in building a complete travel plan within our prototype.

Another important achievement is that this project did not start from assumptions alone. It was inspired by interviews and research with different people, which helped us confirm that this is a real problem many users face when interacting with AI systems.

Finally, we are proud that GUY is designed with the user in mind. Instead of feeling like they are simply giving commands to a machine, users interact with a coordinated team of agents with distinct roles and identities, creating a more comfortable, trustworthy, and human centered experience.

What we learned

Throughout this hackathon, we explored and experimented with several new technologies, particularly Amazon Nova and the broader AWS ecosystem. Working with Nova allowed us to rapidly prototype ideas and test different approaches for building AI-powered solutions.

One of the biggest takeaways was how quickly we could go from concept to working prototype. Tasks that would normally take hours of manual coding or experimentation could be accelerated significantly with AI assistance. This was especially noticeable when iterating on application logic, generating code structures, and refining different components of the system.

We also discovered how powerful the AWS environment can be for building scalable AI applications. Being able to integrate different services and experiment within the platform made it easier to focus on the problem we were solving rather than spending too much time on infrastructure.

Another key learning experience was collaborating efficiently as a team while developing the prototype. Tools such as Visual Studio Code helped us organize the project, manage the codebase, and iterate quickly as we tested new ideas and improvements.

Overall, this hackathon pushed us to explore new AI capabilities, experiment with emerging tools, and rethink how quickly ideas can be transformed into functional prototypes.

What's next for GUY:

Our next step is to expand GUY’s ecosystem of specialized agents. We aim to introduce more expert agents across different domains, allowing GUY to better understand what each agent does and coordinate the right ones depending on the user’s needs.

We also want to give users more control over the process. Instead of only accepting the agents GUY recommends, users will be able to choose their preferred agents, creating a more personalized and transparent experience.

For the travel prototype specifically, we plan to further improve the user experience so it feels less like a simple travel planning tool and more like a complete assistant that users can rely on throughout their journey.

In the long term, our vision is for GUY to evolve beyond travel and become a flexible AI project manager capable of coordinating specialized agents to help users develop many different types of projects.