Telegram bots, with zero-friction setup, are versatile for various tasks such as support chats, weather updates, games, and more.Bots function by listening to the Bot API, processing user input, and generating responses through a web service.Improvements in Telegram bots involve employing large language models (LLMs) for intent recognition from free-form text.AstroGPT, a sandbox project, employs LLMs for astrology advice, registration, and generating horoscopes.AstroGPT's initial implementation used if-else logic for human-like interactions, and it evolved to leverage LLMs for reasoning.The bot's logic is structured into prompts and actions that guide the conversation flow and decision-making for personalized responses.Features like merging extraction of values, menu prompts for action selection, and advice prompts enhance the bot's conversational capabilities.Decision-making in the bot includes selecting next actions, providing advice based on user input, and summarizing results for user replies.Safety measures include spotting abnormal behavior with confidence scores and warning types to prevent abuse in the bot interaction.The use of LLM-reasoning in Telegram bot design improves conversation flow, enables handling complex queries, and enhances the human-like interaction.