Description
Summary We are building a premium, self-hosted B2B SaaS platform designed to automate client intake, onboarding, and inquiry management directly inside Telegram. We are looking for an expert full-stack Python engineer with deep experience in advanced Telegram automation and self-hosted AI workflows to design, build, and deploy our backend architecture. Important Architectural Note: This platform does NOT use the standard Telegram Bot API (BotFather). To maintain an authentic, high-end, human-to-human interaction model, each onboarded client will connect their own dedicated Telegram account via a User Bot (using Telethon or Pyrogram). Their customers will interact with what appears to be a standard, organic Telegram user profile. System Architecture & Core Workflows: Multi-Tenant User-Bot Management: Build a highly secure, isolated PostgreSQL database structure capable of managing 20–25 independent clients simultaneously. Each client profile will run its own autonomous User Bot session. Messages, sessions, and routing keys must be perfectly isolated to prevent cross-client data contamination. Secure Client Onboarding Portal (No-Code View): Build a lightweight, front-end onboarding web form (or integrate a secure open-source form tool like Formbricks or Typebot) that generates a unique link for each new client. Via this link, clients must be able to input their own operational details: pricing/rates, session durations, services, precise address/location data, and custom persona instructions (including their preferred vocabulary, slang, and tone of voice). Submitting this form must automatically update their specific PostgreSQL row and immediately train the chatbot. Crucially, clients must never see the underlying code, database architecture, or Flowise workflows. Automated Onboarding & Persona Execution: Every time a new customer initiates a chat, the User Bot must automatically execute an initial qualifying workflow based on the client's submitted form data. It must instantly deliver the client's specific rates, durations, and services. Following this introduction, the AI must seamlessly handle FAQs and maintain a natural, fluid conversation using the exact vocabulary and slang defined by the client to completely mask the use of automation. AI Engine & Translation Pipeline: Route incoming customer inquiries from the client’s User Bot account through a self-hosted Flowise instance (connected to a centralized Ollama running Llama 3.2 on an unmanaged Linux VPS). The AI's response and conversation transcript must pass through a local translation layer (e.g., LibreTranslate) to match the client's preferred native language before delivery to the client. Private Master Chat & Custom Controls: The formatted, translated transcript must be forwarded by the system to the client's separate, private Telegram account. This master chat will feature custom Inline Keyboards with three actionable controls: [Accept], [Deny], and [Counter Offer]. Sensitive operational data (such as the address or location coordinates inputted during onboarding) must be securely withheld by the system and only released to the customer once the client manually taps [Accept]. Bidirectional "Swipe-to-Reply" Routing: Implement a precise message-ID mapping schema in PostgreSQL. If a client chooses to manually intervene, they will use Telegram's native "swipe-to-reply" feature on any specific transcript inside their private master chat. The Python script must intercept this reply, map it back to the active customer session, and deliver only the raw text of the client's manual response via the User Bot profile. The customer must never see the underlying master chat layout, formatting, or buttons. Strict Data Hygiene (48-Hour Purge): To maintain peak server performance and ensure absolute privacy, build an asynchronous background worker (using pg_cron or an optimized asyncio loop) that executes a hard deletion of all database records and associated Telegram message histories via the Telethon API exactly 48 hours after a chat session becomes inactive. The Tech Stack You Will Use: Language: Python 3.10+ (Asynchronous design is mandatory) Frameworks: Telethon or Pyrogram (for robust User Bot / MTProto session management) Frontend/Forms: A lightweight framework (FastAPI/JinJa2, Next.js, or containerized tool) to handle the secure onboarding link. AI & Workflow: Flowise & Ollama (Llama 3.2 deployed and optimized for CPU multi-threading) Database: PostgreSQL (with pgvector for dynamic context/memory storage) Containerization: Docker & Docker Compose Target hosting: Unmanaged Linux VPS (AlexHost infrastructure)