Telegram AI Sales Bot for a D2C Jewelry Brand

The problem

A D2C jewelry brand was selling through Telegram, and every sale started the same way: a customer asking questions. Metal purity, stone details, sizes, prices, delivery times. The founder was answering these personally, one chat at a time, which meant slow replies during busy hours and no replies at night — and in D2C, an unanswered question is usually a lost sale.

What I built

A Telegram sales assistant that answers like a knowledgeable shop assistant, not a generic chatbot.

• Grounded answers, not internet answers. The bot runs on Gemini Flash, but every reply is grounded on the brand's actual product data. Ask about a specific piece and it answers from the catalog — materials, pricing, availability — instead of hallucinating jewelry facts.
• Voice-note replies. This is the detail customers remember. The bot can reply with a voice note, generated with edge-tts and converted with ffmpeg into Telegram's voice format. In a chat culture where voice notes are how real people talk, a bot that speaks feels personal rather than robotic.
• Sales-aware behavior. The system prompt keeps the bot on-topic: it answers about products, nudges toward a purchase, and hands off to the human owner when a conversation needs one.
• Client-owned deployment. The whole system runs on a Linux VPS under the client's account. Their bot token, their server, their data. I handed over the code with documentation.

Outcome

The brand's Telegram channel answers product questions instantly, around the clock, in text or voice — and the founder stepped out of the role of full-time chat support. The bot handles the repetitive 80% of questions; the humans handle the conversations that actually need them.

Want a bot like this for your store? This is exactly what my agency Telezoid builds — see Telegram bot development.