Inspiration

All four of us have dealt with serious housing maintenance issues. Alp, Henry, and Evan have been dealing with ants for two weeks, and Seth encountered mold in the apartment he moved into. We wanted to build an agent that tackled something we were experiencing ourselves. We also recognized that property managers juggle competing requests, limited budgets, and busy maintenance teams. We built Dispatchr to help both sides: tenants need their problems resolved, and managers need a practical, cost-effective way to get repairs moving.

What it does

Dispatchr is a maintenance platform for tenants, property managers, technicians, and vendors. Tenants describe their issue and answer follow-up questions that help clarify the problem and identify hazards. An agent then checks maintenance history, warranties, staff availability, and local repair prices. The manager receives a dashboard of repair options with estimated costs and explanations, including whether sending a generalist first makes sense or could lead to an expensive second visit. Managers approve the next step, while emergencies bypass cost comparison and escalate immediately.

How we built it

We built the web app with Next.js, React, TypeScript, and Tailwind CSS, with SQLite storing requests, repair plans, approvals, and activity logs. The agent uses Meta’s Muse model to interpret requests and call tools that retrieve property records, check repair resources, and research prices online. We kept safety rules, qualification checks, cost calculations, and approval requirements in application code. That lets the model investigate while the application enforces the boundaries. We used Vitest for regression tests and built an evaluation harness to compare plain model responses with Dispatchr’s research and planning workflow.

Challenges we ran into

Figuring out how to evaluate the system was one of the hardest parts. A convincing recommendation does not automatically mean a repair plan is correct, safe, or useful. We had to distinguish researched prices from estimates, account for uncertain diagnoses, and test what happens when quotes exceed a budget or vendors fail to respond. We also worked to keep that complexity understandable in the interface so managers could make a decision without digging through every agent action.

Accomplishments that we're proud of

We’re especially proud of the attention we gave testing and evaluations. In our recorded comparison, Dispatchr completed researched plans for all three test scenarios, priced five to seven feasible options per case, and prepared each plan in roughly 29–50 seconds. These tests used fictional property records and live web research, demonstrating plan preparation rather than proven customer savings. We’re also proud of the agent workflow and the user experience around it. A tenant starts with a simple description, and a manager gets clear options with supporting evidence. We focused on getting the small things right, including follow-up questions, readable cost breakdowns, and visible approval steps.

What we learned

We learned that building an agent and proving it works are separate challenges. Evaluations take careful thought, especially when the output combines uncertain diagnoses, researched information, and operational decisions. As aspiring forward-deployed engineers, we need to show where a system helps, where it fails, and what still needs human judgment. We also learned how much thoughtful interface design matters when turning a complicated workflow into something people can actually use.

What's next for Dispatchr

Our next priority is connecting approval to real execution. After a manager approves a pest-control plan, for example, we want the agent to call the provider, confirm pricing and availability, and coordinate the appointment within the approved limits. Vendor communication is currently simulated in the demo. We also want to test Dispatchr with real property managers and maintenance requests, measuring resolution time, repeat visits, and actual repair costs. Longer term, we’d bring the workflow into existing maintenance platforms and email systems so teams can use Dispatchr alongside the tools they already know.