Smart Support Chat (Chat Assistant)

What it is

Smart Support Chat is a customer support platform built from scratch. It pairs a rule based chat assistant with real time human chat, and both arrive on a website through the same embeddable widget.

It is worth being precise about the assistant: it is not an AI model. Replies are driven by keyword rules that are defined in the dashboard, which makes its behaviour predictable and cheap to run. For the common questions that make up most support volume, a rule that always answers correctly beats a model that usually does.

How a site gets it

A business registers, configures the domain the widget will run on, and gets a token from its dashboard account. Dropping a small JavaScript snippet carrying that token onto the site puts a working floating chat window on every page. There is nothing else to install and no framework requirement, which keeps it usable on sites the support team does not control.

The handoff to a human

Rules only carry a conversation so far. When a visitor types something the rules do not cover, such as asking for support directly, the conversation is escalated: a request goes to the human agents logged into the system, and one of them picks it up and continues the same thread in real time. The visitor is not asked to repeat themselves or move to email.

That escalation path is the reason the platform has real accounts and roles rather than just a script. Agents are users of the system, so a business can staff the widget during working hours and let the rules cover the rest.

How it works

Laravel and MySQL with WebSockets carrying the live messages, and Tailwind CSS for the dashboard and widget. Each assistant is scoped to its domain and token, so one installation cannot read another's conversations.

My role

About two months, built from scratch: the data model, the backend, the dashboard where rules are written, and the embeddable widget itself.

See it working

The dashboard and the agent side sit behind an account, so the parts worth seeing are not public. This walkthrough shows the assistant, the handoff to a human and the dashboard: Demo Link.