FixPilot

FixPilot is an autonomous property-maintenance agent for small property owners.

Problem

Owners lose time reading tenant reports, judging urgency, finding vendors, comparing quotes, and arranging routine work.

Solution

A tenant report becomes an incoming event. FixPilot classifies the issue, searches local vendors, requests and compares mock quotes, and schedules the selected repair automatically when it is $150 or less. Higher-cost work is held for an owner approval or rejection in the monitoring dashboard.

How AI is used

The FastAPI service creates a real Strands Agent with five native tools: classify_request, search_vendors, request_quotes, compare_quotes, and schedule_repair. The validated live configuration uses Amazon Bedrock with Amazon Nova 2 Lite.

In a live run, the agent used all five tools for a $125 air-conditioner repair and scheduled it automatically. For a $600 water-heater repair, it stopped after quote comparison and surfaced an owner approval request. Ollama is an optional local fallback; deterministic demo mode preserves the same workflow and policy.

Key features

• Tenant Portal that emits a simulated maintenance event
• Owner dashboard for Agent Activity, Active Requests, and Human Approval
• Automatic scheduling at or below $150
• Human approval only when the selected quote exceeds $150
• Local mock vendor, quote, and scheduling data; no vendor, email, or SMS service is contacted

Testing

Run the FastAPI backend and Next.js frontend locally. Submit a low-cost air-conditioner leak to confirm automatic scheduling, then a water-heater failure to confirm the approval boundary. Backend tests, lint, typecheck, and production build pass locally.

Limitations

Vendor discovery, quotes, bookings, email, and SMS are local mocks. Requests and approvals are in memory and do not survive an API restart.

Built With

• amazon-bedrock
• amazon-nova-2-lite
• fastapi
• next.js
• python
• strands-agents
• typescript