About the project
We built a high-load, intelligent Telegram bot designed to automate sales workflows, qualify leads, and support users for education products.
The solution is a hybrid system that combines a classic scripted sales funnel (powered by a Finite State Machine) with generative AI (LLM) to enable natural, personalized dialogue with prospects.
The bot acts as the first line of the sales team: it operates 24/7, handles inbound requests, nurtures the audience, and routes “hot” clients to managers.
User-facing functionality (Client)
For the end user, interacting with the bot feels like speaking to a competent consultant who understands context and helps choose the right solution — not just a set of canned replies.
Natural dialogue powered by AI
The user asks questions in free form. Using LLM, the bot understands intent, answers product‑specific questions, handles objections, and keeps context across the conversation.
Personalized sales funnel
The bot guides the user through funnel stages: discovery and needs analysis to product presentation and closing. The flow adapts to user responses with relevant content.
Buffering and smart replies
Debounce/Buffering: when a user sends multiple short messages in a row (e.g., "Hi", "How much?", "Any installments?"), the bot avoids spam and responds once with a single, comprehensive answer.
Access to materials and bonuses
Users can receive useful files, guides, and presentations right in chat—on demand or as rewards for completing questionnaire steps.
Multimedia support
The bot handles not only text but also images with captions and voice messages (with transcription), making communication convenient for users.
Functionality and benefits for the company
For business, the bot scales sales, reduces manager workload, and increases conversion through instant responses and high‑quality qualification.
Automatic Lead Scoring
The system analyzes every user message and assigns a "warmth" score using heuristics and AI semantic analysis. This enables segmentation and directs sellers’ attention to the most promising leads.
Intelligent lead handoff
The bot decides when to transfer the dialog to a live manager: upon high score, explicit user request ("want to call"), or complex edge cases. A lead card with history and context is instantly sent to the admin channel or CRM.
Full-featured CRM inside Telegram
- Admin panel: Manage products, pricing, and content without code edits.
- Request management: Managers pick up requests, change statuses, view client profiles, and message them directly via the bot UI.
- Notifications: Instant alerts for new leads and user actions.
Powerful marketing toolkit
- Broadcasts: Launch segmented mass messaging, including A/B testing for titles and content.
- Follow‑up chains: Automatic reminders to re‑engage users who went silent for 24/72 hours.
- Context awareness: All automated messages are logged into the dialog history so AI "remembers" what was sent and can continue the discussion.
Analytics and reporting
Built‑in metrics collection (users, conversions, sales, broadcast efficiency) with data export capability.
Technologies and Integrations
The project is built on a modern tech stack that ensures high performance, resilience, and security.
Backend
Python 3.11+, FastAPI (for webhook and metrics API), aiogram 3.x (async framework for Telegram).
Artificial Intelligence
OpenAI API integration (GPT‑4/GPT‑4o mini) for answer generation and semantic analysis. RAG (Retrieval‑Augmented Generation) is used to query the product knowledge base.
Databases
- PostgreSQL: Primary storage (users, messages, transactions, leads). SQLAlchemy 2.0 and asyncpg are used for async access.
- Redis: Caches FSM states, stores temporary message buffers, enables idempotency (duplicate protection), and coordinates distributed tasks.
Infrastructure
Architectural features
- FSM (Finite State Machine): Clear control over funnel states.
- Idempotency Middleware: Protects from duplicate update processing on network glitches.
- High Concurrency: Tuned connection pools and workers support 1000+ concurrent active dialogs.
Architecture & components
The project is split into several clear layers (Clean Architecture):
Implementation results
Implementation created an autonomous sales channel that efficiently processes inbound traffic.
Routine automation
The bot handles 100% of first contacts, filters out low‑quality traffic, and answers FAQs — freeing managers for key clients.
Faster response time
Response time reduced to seconds — critical for keeping attention in messengers. Buffering made bot replies more human‑like and complete.
Uninterrupted operation
Thanks to a microservice architecture and queues, the bot remains stable under load, without losing messages even during marketing peaks and mass broadcasts.
Process transparency
The client gained full control over the sales funnel via detailed dialog logging and transparent analytics, tracking script effectiveness and A/B tests in real time.