Left On Bread.

Inspiration

What started as an attempt to optimize the shopping experience at this hackathon turned into a four-hour session designing an algorithm for one customer. Somewhere along the way, we realized that shopping is often a team decision.

As college students, we constantly face the challenge of buying things for shared apartments. From furniture and kitchen appliances to everyday essentials, different preferences, budgets, and priorities can turn simple purchases into endless group chats and difficult compromises.

This inspired us to build Left on Bread, a collaborative shopping platform where AI agents help roommates reach a consensus without the endless back-and-forth. Each person decides what they want, what matters to them, and how much they're willing to spend. Their AI agents then advocate for their needs, compare trade-offs, and help the group find an option that works for everyone.

We wanted to use AI to make decisions together easier while keeping people in control of their choices. Because when you're juggling midterms, deadlines, and everyday life, figuring out what to buy with your roommates shouldn't become another group project.

What it does

Left On Bread is a collaborative shopping website where each individual uses AI agents to recommend products, negotiate based on each person's private budget preferences, and vote on purchases. A purchase can proceed to checkout only when enough members agree, helping households make group buying decisions without endless debates.

How we built it

We started with an algorithm designed to optimize shopping for a single person, but quickly realized the bigger challenge was finding products that worked for an entire group. Inspired by emerging agentic AI systems, including experiments with AI agents negotiating and resolving conflicts with one another, we explored how agents could advocate for individual preferences while learning to compromise toward a shared decision.

We began by designing the core negotiation algorithm and building a working skeleton that allowed agents to communicate, evaluate trade-offs, and work toward consensus. Once the algorithm was functional, we built the skeleton of our website and integrated the AI agents into the codebase, connecting the negotiation process to the user experience. We then extensively tested different scenarios, identified edge cases and bugs, and iteratively refined the agents' behavior to make negotiations more reliable. Finally, we focused on CSS and UI improvements to create a cleaner, more intuitive experience for users.

Challenges we ran into

One of our biggest challenges was deploying our website, from connecting our GoDaddy domain to our homelab server to configuring Cloudflare Tunnels for secure external access. Beyond deployment, we worked through integrating our AI agents into the website and correctly assigning each agent its role within the codebase. Since different AI engines helped build different parts of the project, debugging inconsistencies and refining targeted prompts required significant iteration. We also spent considerable time refining the CSS, making the UI intuitive to navigate, and implementing a reliable database and data engine.

Accomplishments that we're proud of

We're proud of turning our idea into a working website, bringing the entire experience together into a functional product. Hosting the project on our own homelab also gave us greater flexibility and control over server-side customization. Most importantly, we successfully integrated AI agents that actively advocate for individual preferences and negotiate with one another to help the group reach a decision.

What we learned

Throughout this hackathon, we gained hands-on experience with how agentic AI systems operate, including how agents process information, make decisions, and negotiate within surprisingly short execution windows. We also learned how to systematically debug complex AI behaviors and use prompt engineering to make agent interactions more reliable, efficient, and aligned with their intended roles.

What's next for Left On Bread

We want to continue developing Left On Bread into a polished, practical product that people can actually use in their everyday lives. This means exploring a dedicated mobile app, improving our server infrastructure, and potentially introducing distributed systems to support more users and concurrent agent interactions. We also hope to strengthen our networking and cybersecurity foundations as the platform grows, building toward a reliable, secure, and scalable collaborative shopping experience.