Telegram bot for project technical support. Answers user questions based on uploaded documentation, using a RAG pipeline (knowledge base search + LLM).
Work channels:
Direct messages — full dialogue with history and context, detailed responses up to 1500 characters
Group chats — automatically detects questions using heuristics (question words, ?, keywords for errors/help), replies with a response; maintains dialogue context through reply-chain and history of recent messages
Inline mode — request from any chat
Stack:
aiogram 3 — Telegram Bot API
ChromaDB + sentence-transformers (E5 multilingual) — vector search through documentation
OpenRouter API (gpt-4o-mini) — response generation
SQLite (aiosqlite) — message history, FAQ candidates, document metadata
Key functionality:
Uploading documentation in formats .txt, .md, .pdf, .docx, .html — chunking, embedding, storage in ChromaDB
Semantic search with relevance thresholds (0.72 for private messages, 0.78 for groups)
Anti-hallucination — prompt requires verbatim citation of documentation, prohibits logical conclusions
Automatic compacting of long dialogues through summarization
Collection of FAQ candidates (frequently asked questions)
Admin commands
/list_docs, /delete_doc, /stats, /export_logs, /compact, /faq_candidates
#chat-bot #Telegram-developer #Telegram #AI #api-development #pyTelegramBotAPI #telebot #pytelegrambot #python #Python-developer #Databases #TestDocumentation
Work channels:
Direct messages — full dialogue with history and context, detailed responses up to 1500 characters
Group chats — automatically detects questions using heuristics (question words, ?, keywords for errors/help), replies with a response; maintains dialogue context through reply-chain and history of recent messages
Inline mode — request from any chat
Stack:
aiogram 3 — Telegram Bot API
ChromaDB + sentence-transformers (E5 multilingual) — vector search through documentation
OpenRouter API (gpt-4o-mini) — response generation
SQLite (aiosqlite) — message history, FAQ candidates, document metadata
Key functionality:
Uploading documentation in formats .txt, .md, .pdf, .docx, .html — chunking, embedding, storage in ChromaDB
Semantic search with relevance thresholds (0.72 for private messages, 0.78 for groups)
Anti-hallucination — prompt requires verbatim citation of documentation, prohibits logical conclusions
Automatic compacting of long dialogues through summarization
Collection of FAQ candidates (frequently asked questions)
Admin commands
/list_docs, /delete_doc, /stats, /export_logs, /compact, /faq_candidates
#chat-bot #Telegram-developer #Telegram #AI #api-development #pyTelegramBotAPI #telebot #pytelegrambot #python #Python-developer #Databases #TestDocumentation