AI consultant for tire selection in Telegram
Task
The tire shop is losing inquiries during non-working hours and wasting managers' time on standard questions: "which tires are suitable for my car," "how much do they cost," "are they in stock." A system is needed that conducts initial consultations 24/7, qualifies clients, and passes warm leads to the manager.
Solution
A Telegram bot based on LLM, orchestrated through n8n. The bot operates like a live consultant: it determines the client's language, collects parameters in a natural dialogue format, selects suitable tires from the real store catalog, and guides the interested client to a request — with contacts and a list of desired models in a table for the manager.
Architecture
Telegram Trigger → AI Agent (GPT-4o-mini) with dialogue memory and access to the tire catalog (Google Sheets as a tool) → parsing the response and extracting structured data → sending to the client + logging → merge logic for leads → CRM table.
Key Features
Smart Dialogue
Auto-detection of language (Ukrainian / Russian) from the first message and maintaining it throughout the dialogue
Memory for 20 messages — the bot remembers the context
Funnel: car parameters → season → budget → selection → processing
Short messages with 1-2 questions at a time — like a live manager, not like a survey
Selection from the Real Catalog
The agent calls Google Sheets as a tool, filters by size and season
Provides 2-3 options with prices, availability, and a brief justification for the choice
Does not hallucinate: if the product is not available — it honestly states so and offers an alternative
CRM Funnel
Classification of each dialogue: HOT / WARM / COLD
Hot leads are saved in a separate table with contacts and a list of recommended tires
Automatic calculation of conversion time (from the first contact to readiness to buy)
Status model: New → Hot → Contacted → Closed — the manager leads the client in the same table, the bot does not overwrite their changes
Analytics
All dialogues are logged with reference to the lead status
Data for optimization: what inquiries come in, at what step clients drop off, which tires are recommended more often
Scalability
The current architecture is optimal for catalogs of up to 300-500 items — the full catalog is loaded into the LLM context, ensuring maximum accuracy in selection. For larger catalogs, it can easily scale to a RAG architecture with vector search (Qdrant / Supabase).
Result
The bot completely replaces the manager at the initial qualification stage. The manager receives only hot clients with ready information: what they want, which tires they viewed, when they contacted, and their contacts. The lead processing time is reduced significantly.
Stack
n8n · OpenAI API (GPT-4o-mini) · Google Sheets API · Telegram Bot API · JavaScript (Code nodes) · Prompt Engineering
The tire shop is losing inquiries during non-working hours and wasting managers' time on standard questions: "which tires are suitable for my car," "how much do they cost," "are they in stock." A system is needed that conducts initial consultations 24/7, qualifies clients, and passes warm leads to the manager.
Solution
A Telegram bot based on LLM, orchestrated through n8n. The bot operates like a live consultant: it determines the client's language, collects parameters in a natural dialogue format, selects suitable tires from the real store catalog, and guides the interested client to a request — with contacts and a list of desired models in a table for the manager.
Architecture
Telegram Trigger → AI Agent (GPT-4o-mini) with dialogue memory and access to the tire catalog (Google Sheets as a tool) → parsing the response and extracting structured data → sending to the client + logging → merge logic for leads → CRM table.
Key Features
Smart Dialogue
Auto-detection of language (Ukrainian / Russian) from the first message and maintaining it throughout the dialogue
Memory for 20 messages — the bot remembers the context
Funnel: car parameters → season → budget → selection → processing
Short messages with 1-2 questions at a time — like a live manager, not like a survey
Selection from the Real Catalog
The agent calls Google Sheets as a tool, filters by size and season
Provides 2-3 options with prices, availability, and a brief justification for the choice
Does not hallucinate: if the product is not available — it honestly states so and offers an alternative
CRM Funnel
Classification of each dialogue: HOT / WARM / COLD
Hot leads are saved in a separate table with contacts and a list of recommended tires
Automatic calculation of conversion time (from the first contact to readiness to buy)
Status model: New → Hot → Contacted → Closed — the manager leads the client in the same table, the bot does not overwrite their changes
Analytics
All dialogues are logged with reference to the lead status
Data for optimization: what inquiries come in, at what step clients drop off, which tires are recommended more often
Scalability
The current architecture is optimal for catalogs of up to 300-500 items — the full catalog is loaded into the LLM context, ensuring maximum accuracy in selection. For larger catalogs, it can easily scale to a RAG architecture with vector search (Qdrant / Supabase).
Result
The bot completely replaces the manager at the initial qualification stage. The manager receives only hot clients with ready information: what they want, which tires they viewed, when they contacted, and their contacts. The lead processing time is reduced significantly.
Stack
n8n · OpenAI API (GPT-4o-mini) · Google Sheets API · Telegram Bot API · JavaScript (Code nodes) · Prompt Engineering
Odessa 
Available for hire
Offer work