Product · 2024
An AI support layer that sounds like the brand, not the vendor.
Chatloom is the AI chat product I built for teams drowning in repetitive support tickets. Faster answers, fewer humans on "did you try turning it off".

Role
Product design & build
Year
2024
Stack
Next.js, OpenAI, Supabase
Scope
Product UI, admin, marketing
Context
Every AI chatbot on the market sounds the same; the same polite robotic preamble, the same "I'd be happy to help with that," the same sterile gray bubble. Customers smell it. It breaks trust before the conversation starts.
I wanted Chatloom to feel like it was part of the brand that deployed it. Configurable voice, configurable tone, configurable refusal behaviour; and a product surface that isn't another SaaS dashboard.
The challenge
The challenge
AI chat is an easy category to enter and a hard category to stay in. I wasn't trying to beat the big names on raw model quality; I was trying to beat them on feel, setup speed, and how little a small team has to wrestle with it. Product design as positioning.
How I built it
A product with a personality.
A chat surface that respects the site
Most chatbot widgets stamp a bright logo over your layout. Chatloom inherits the host site's type and radius, reads the accent color from the page, and lives in a panel shaped like it belongs there. It stops being a widget and starts being a room.

Voice is a first-class setting
Instead of burying tone controls in an advanced panel, voice is the first thing you configure. Formal, friendly, deadpan, playful; with live preview, not abstract sliders. You write one example message in your brand tone and Chatloom mirrors it from there.
Refusal as a designed moment
Most chatbots fail the interesting moment; when they don't know the answer; with a scripted "I can't help with that." I designed the refusal flow as a first-class experience: graceful handoff, captured context, a real human picks up where the bot stopped. No dead ends.

Setup in an afternoon
No SDKs, no ticket to a solutions engineer. Paste a script tag, upload a PDF or a URL, answer five questions about tone. You're live before lunch ends. I design every onboarding screen as if the user is about to give up.
Product surfaces
The product, at rest.


What I shipped
Deliverables
Outcome
A chat product that stops feeling like a chatbot.
Chatloom landed with teams who were tired of bolt-on AI widgets. The voice configuration gets used on day one; the handoff flow gets used in week one; and the support load drops without the support team having to apologize for a robot. It's the difference between AI that's deployed and AI that's adopted.
Personality ships faster than features.
WANT YOUR OWN?
Need an AI layer that sounds like your brand?
Start with a free Revenue Check-up. We figure out where AI actually helps, together.
Next project