FleetMind · AI Operating System for Physical-Service Operations · Build with Gemini XPRIZE

Important

🏆 GRAND PRIZE POOL: $2,000,000 USD (2 Million Dollars!)

1st Place: $500,000 USD! Let's build the state-of-the-art AI-native physical operations platform and bring home the victory! 🌟🚀

Forenly's entry for the Build with Gemini XPRIZE — build a business that operates with AI.

FleetMind is an AI "Ops Autopilot" that runs the day-to-day operations of physical-service businesses end-to-end. It takes the job, schedules the crew, dispatches them, confirms the work is done, and sends the invoice — a zero-human ops loop powered by Gemini + Google Cloud. Humans set goals and handle exceptions; the agents run the loop.

First vertical: commercial cleaning (where we earn real revenue). Robot fleet = the autonomous-future demo layer — the same ops loop where a floor-scrubber reports completion telemetry instead of a human tapping "done".

Community

Building in the open — team chat: https://discord.gg/vJpBqa9C7q

What the business does

Small cleaning companies (3–25 cleaners) run their operations on WhatsApp + spreadsheets. FleetMind replaces that with an agent layer that runs the whole loop:

Agent	Role	Gemini capability
Intake	Turns a free-text/photo request into a structured job	Structured output + vision
Scheduling	Assigns crew + time slot under real constraints	Multi-step planning + function calling
Dispatch	Texts the crew the job + checklist; re-plans on no-show	Function calling (Twilio/WhatsApp)
Verification	Checks the completion photo against the job scope	Vision
Billing	Generates the invoice from the completed-job log; handles disputes	Structured output (→ Stripe)

Revenue is charged to real third-party cleaning companies for the work the agents run — meeting the XPRIZE arms-length-revenue requirement directly.

Why this fits "build a business that operates with AI"

There are no human operators scheduling jobs or writing invoices. Gemini is in the critical path of every operational decision — the AI-native-operations criterion, literally. The robot fleet demo shows where the same loop goes next (autonomous crews).

⚠️ This is not a code hackathon — it's a real business

Judging is on a live, revenue-generating business run during the 90-day window. Three equally-weighted criteria:

1. Business Viability — real users, real arms-length revenue, sustainable model.
2. AI-Native Operations — AI live in production making the key operational decisions.
3. Category Impact — meaningfully moves the needle.

Category

• Small Business Services ← giving everyday service businesses an AI back-office
• Education & Human Potential · Entrepreneurship & Job Creation · Money & Financial Access · Professional Services

Two hard eligibility constraints

1. New project only — created after 19 May 2026. Boilerplate allowed but must be explained.
2. Arms-length revenue only counts. Revenue from team/family/related/pre-existing customers is reported separately. The business must serve real third-party customers.

Required stack

• At least one Google Cloud product — Cloud Run (orchestrator + portal), Firestore, Vertex AI.
• Gemini API for ≥1 LLM call in the deployed app — the agent decisions above.

Architecture

Public landing + owner portal (Vue · Cloud Run)
        │
        ▼
  Agent Orchestrator (Gemini API · FastAPI · Cloud Run)
  ├── Intake → Scheduling → Dispatch → Verification → Billing   (the zero-human ops loop)
        │
        ▼
  Firestore (clients · crews · jobs · invoices · agent_events)
        │
        ├── Twilio / WhatsApp  (dispatch the field crews)
        ├── Stripe             (real invoices → real revenue)
        └── Robot fleet simulator  ← DEMO / VISION LAYER (simulator/)

See docs/MVP-ARCHITECTURE.md (the loop + Week-1 ship cut), docs/GTM-ACQUISITION.md (first-customer plan), and portal/ (the Vue operations portal).

Key dates

Date	Event
19 May 2026	Submission opens (project must start after this)
17 Aug 2026, 13:00 PT	Submission deadline
18 Aug – 15 Sep 2026	Judging
~25 Sep 2026	Finalist pitch + winners

Submission deliverables

• Devpost page + category · public repo (license) or shared with testing@devpost.com + judging@hacker.fund
• Text description · demo video <3 min showing the product running
• Revenue evidence (total + by month; related-party separate) · cost evidence · user evidence
• Production-running evidence — agent execution logs, API usage, dashboard screenshots

Prizes

1st $500k · 2nd $200k · 3rd–5th $100k · 15× runner-up $50k · 5× category $50k. Max one prize per project.

Status

🚧 Direction locked: AI Ops Autopilot for commercial cleaning (robot fleet = demo layer). Building the Week-1 revenue slice (intake → schedule → dispatch → verify → Stripe invoice).

Team

Forenly · github.com/Forenly · Apache-2.0