AI builder who teaches what he ships — new production AI lessons
    Jasmin.
    All Architectures
    AI SystemsAdvanced

    AI Support Desk

    A support stack where the bot answers from your real tickets and docs (pgvector), escalates with full context, and sits beside human agents as copilot. Redis holds the live conversation and draft suggestions. Humans stay in the loop for refunds and angry customers.

    500 – 30Kusers supported
    $250 – $2,200/month infrastructure

    Architecture Diagram

    Interactive — hover over any node to see its role and description.

    CustomerHuman AgentInbox UIAssist Orch.LangGraphpgvectorDocs + ticketsPostgreSQLTicketsRedisLive threadLLMUserFrontendBackendDatabaseCacheAI / ML

    Use Cases

    Deflecting tier-1 questions 24/7
    Agent copilot that drafts replies from past tickets
    Unified view of chat, email, and WhatsApp
    Policy-grounded answers with citations
    Handoff to CRM with a clean transcript

    Technology Stack

    frontend

    Next.js inboxwidget

    backend

    Node.jsLangGraphFastAPI

    database

    PostgreSQL + pgvectorRedis

    ai

    GPT-4oembeddings

    integrations

    Zendesk / custom ticketsSlack

    Scalability Roadmap

    Stage 1One brand users· Bot + pgvector

    Widget + ingest help centre. Human email fallback.

    Stage 2Team inbox users· Copilot + Redis

    Agents in one inbox. Redis for live threads. Macros.

    Stage 3Multi-brand users· Namespaces

    pgvector per brand. Shared orch. Reporting warehouse.

    Stage 4Global users· Regional desks

    Data residency for tickets. Redis local. Model routing by language.

    Cost Breakdown

    Development Cost

    $14,000 – $35,000 (8–16 weeks)

    Infrastructure Cost

    $250 – $2,200/month (LLM is the variable)

    Maintenance Cost

    $1,500 – $4,000/month for evals, macros, and knowledge hygiene

    Security Considerations

    Customers only retrieve their tenant’s docs and their own tickets
    Refunds and account changes require a human tool confirmation
    Transcripts retained per policy; PII redaction optional
    Widget tokens scoped; no admin MCP from the public chat

    More Architectures

    AI Systems

    AI Agent Platform

    Autonomous AI agents with memory, tools & multi-step reasoning

    AI Systems

    RAG Chatbot System

    AI chatbot grounded in your private documents and knowledge base

    AI Systems

    AI Voice Assistant

    Real-time AI voice calls with STT, LLM reasoning, and TTS response

    Need This Architecture Built?

    Get a detailed architecture plan, technology recommendations, development roadmap, and infrastructure estimation for your project.