AI that connects conversations to lead generation.

Problem

Most website chatbots either answer generic questions badly or dead-end the visitor. The valuable moment — a visitor asking something the site does not cover — is exactly when a business wants a human to follow up.

Context

A Flask application deployed behind a WSGI server, running on bkclassroom.com.

Architecture

Website content grounds the model's answers; when a question falls outside that context, the conversation switches to lead capture and notifies the owner.

Engineering decisions

Ground answers in website context

Keeps the assistant on-topic for the business rather than acting as a general chatbot.

Treat out-of-scope questions as leads

The question the bot cannot answer is a sales signal, so it becomes a capture step instead of a failure.

Telegram for notifications

Owners get an alert on their phone while the visitor is still engaged.

How clients can see the product

The admin dashboard is not a public demo — it holds other people's bots and leads. Clients get a public chat preview, this case study, and a live walkthrough on a call. Do not share dashboard logins.

• Public preview: https://bkclassroom.com/chat/leadGenie1 — visitor chat, booking prompt, transcript download.
• This page: screenshots of the dashboard, create form and live site.
• Live call: create a bot, embed it, and watch a lead land in Telegram.

What the operator can do

After login, the dashboard lists each chatbot as a card. From there you preview the widget, edit setup, copy embed code, or open that bot's leads. Creating a bot asks for a website ID, URL, a scan of the site, products, contact details, accent colour, logo, Telegram token and optional booking link.

Lessons

• AI is most useful when it hands off to a human at the right moment.