Inspiration

Two secretaries and an office administrator at Cool & Heat Air Conditioning were spending nine hours a day glued to phones, calendars, and inboxes just to keep up with customer requests. We built Joy AI to give that time back.

What it does

Joy AI is a Gemini-powered agent that answers real customer messages for Cool & Heat Air Conditioning — through website chat and the contact form — replies in the business's voice within seconds, and extracts structured booking data (name, phone, address, urgency) to decide on its own whether a request is ready to be scheduled. A live operations panel shows every request the moment it arrives.

How we built it

A PHP proxy calls Gemini (gemini-3.5-flash-lite) for both the conversational reply and the structured extraction in a single request. Every call is logged in production, and each log line is mirrored in real time through Cloud Run into Cloud Storage, so the system's behavior is auditable end to end.

Challenges we ran into

Early iterations relied on other providers (Zavu for SMS, then Anthropic for the contact form) that hit real-world limits — cost, credits, and provider availability. We migrated the entire pipeline to Gemini to keep the system reliable and affordable for a small business client.

Accomplishments that we're proud of

This isn't a demo — Cool & Heat is a paying client, and this is one of eight small businesses currently running on this model. Real customers have been served through it, including specific HVAC repair requests captured and routed automatically.

What we learned

That small businesses don't need a call center — they need an AI operations partner that respects the boundaries the owner sets (never inventing prices, always deferring urgent cases to a human).

What's next

Packaging this playbook so every new small-business client becomes a teachable job for someone else, not just a subscription.

Built With

Share this project:

Updates

Submission history