Inspiration We wanted to build something that solves a real, everyday problem for a large market. Property management felt like the perfect space because real estate involves constant coordination between tenants, landlords, brokers, vendors, and maintenance teams.
The idea clicked when we imagined someone managing multiple properties or closing many tenant deals, then suddenly having to handle complaints, vendor calls, approvals, scheduling, and follow-ups all at once. That workflow can get messy very quickly, especially when plumbers, locksmiths, cleaners, electricians, and other third parties are involved.
That is when Property Care Agent started to feel exciting: an agent that receives a tenant complaint, understands the issue, recommends the right action, asks the manager for approval, communicates with the right parties, coordinates scheduling, and keeps humans involved for sensitive or ambiguous decisions.
What it does Property Care Agent helps property managers handle maintenance requests from complaint to approved action. It analyzes tenant complaints, classifies the issue, checks related tickets, recommends vendors, drafts messages, asks for manager approval, handles vendor replies, and helps schedule work orders.
The agent does not blindly act on behalf of the manager. It keeps approval gates in place before contacting vendors or finalizing work, making it useful while still being safe and auditable.
How we built it We built the project using Gemini API, MongoDB, Codex, a Python agent workflow, a backend API, and a frontend dashboard.
Haseeb, Danial, and Ali worked on the agentic workflow, including complaint analysis, vendor ranking, approval handling, and vendor reply logic. Siddharth integrated MongoDB and connected the agent workflow into the backend and frontend experience.
MongoDB stores the maintenance tickets, tenant records, vendor data, approvals, messages, work orders, and agent action logs. The frontend lets a manager review the recommendation, approve actions, simulate vendor replies, and finalize work orders.
Challenges we ran into One of the hardest parts was designing the vendor scoring system. We had to make sure the agent picked the right vendor based on issue type, emergency availability, property preference, rating, workload, and tenant availability.
Another challenge was making sure Gemini received enough context without giving it too much irrelevant information. We also had to make the prompts strict enough so the model returned useful structured output and did not take unsafe actions without approval.
Accomplishments that we're proud of We are proud that the project became more than a simple one-call chatbot. It acts more like a real workflow agent: it reasons through the complaint, gathers evidence, recommends an action, waits for approval, reacts to vendor replies, and records the process.
We are also proud of the approval-gated design because it makes the agent practical for a real property management setting where mistakes can affect tenants, owners, and vendors.
What we learned We learned how different agentic workflows are from basic LLM calls. A useful agent needs context, state, tools, approvals, memory, and clear boundaries. It is not just about generating a response; it is about moving a real process forward step by step.
We also learned how important structured data is. MongoDB helped us ground the agent’s recommendations in actual records instead of making the system feel like a generic chatbot.
What's next for Property Care Agent Next, we would expand Property Care Agent into a fuller property operations platform. We would add real vendor messaging, calendar integrations, tenant photo uploads, cost approval flows, emergency escalation, and richer analytics for recurring building issues.
The long-term goal is to help property managers reduce manual coordination while still keeping control over important decisions.
Log in or sign up for Devpost to join the conversation.