FRENDT Enterprise-level Telegram AI Bot: Design and Implementation Analysis of Dual-Mode Architecture

This article deeply analyzes the Telegram AI bot project developed by FRENDT. Its core lies in the design of a dual-mode architecture (client mode and internal employee mode), integrating functions such as OpenAI large language models, Google Workspace automation, and HR process automation to build an enterprise-level intelligent solution serving both external customers and internal employees. The project covers scenarios like customer support, case management, HR processes, and internal knowledge base Q&A, achieving end-to-end business automation through multi-system integration.

Amidst the wave of enterprise digital transformation, the integration of instant messaging platforms and AI is reshaping enterprise interaction methods. FRENDT's Telegram AI bot project aims to deeply integrate large language model capabilities into enterprise scenarios, creating a dual-mode intelligent system that serves both external customers and internal employees. This project is not just a chatbot; it is a complete enterprise automation solution covering customer support, case management, HR process automation, and internal knowledge base Q&A, achieving end-to-end automation through the combination of OpenAI GPT, Telegram Bot API, Google Workspace, and PostgreSQL.

Client Mode面向外部客户，提供智能对话（含语音转录）、服务案例管理（收集信息、AI预分析、存储到Google Drive/Sheets）、电缆识别（视觉分析+产品目录匹配）、交互式菜单导航等功能。

Internal Employee Mode通过手机号验证访问，提供HR流程自动化（请假申请/审批/提醒、薪资查询、反馈看板）、实习生任务管理、内部AI助手（文本/文档/图像生成）、通知提醒系统等功能。

Core Technology Stack: Based on Python 3.12, relying on libraries like python-telegram-bot, openai, psycopg2-binary, gspread, pypdf, etc.

Project Structure: Modular design, including entry files, configuration, handlers, Google service integration, GPT assistant, database operations, and other components.

Key Implementations: Contact gateway mechanism (collecting and storing mobile phone numbers); multi-modal AI case analysis (text + GPT-4V image analysis to generate diagnostics); cable identification (visual analysis + product catalog matching); Wizard mode for HR leave process (step-by-step information collection, approval routing, document generation).

Deployment: Supports Docker containerization, configuring Google service accounts, API keys, etc., via environment variables; supports polling or Webhook mode.

Environment Variables: Need to configure Telegram Bot Token, OpenAI API Key, Google service configuration, PostgreSQL connection string, etc.

Security: Need to protect API keys (regular rotation), internal employee mode access control (mobile phone number verification), customer data compliance, and input validation to prevent injection attacks.

The FRENDT project demonstrates typical patterns for enterprise-level AI applications: multi-mode design (differentiated services), LLM integration strategy (scene-based selection of model capabilities), business process automation (embedding into existing processes), and ecosystem integration (deep integration with Google Workspace, etc.).

Insights for developers: This project provides reference implementations in complex interaction processes, multi-modal input, and enterprise system integration, which are worth learning from for similar systems.