Skip to content
Case study · AI Agents

Nexus Support

A multi-agent AI customer support system that routes every ticket to the right specialist, looks up real orders and masks sensitive data.

Nexus Support shown on desktop, tablet and mobile

Overview

Project overview

Nexus Support is a customer support agent built as a team of specialists rather than one general chatbot. A supervisor reads each message and hands it to an orders, technical or billing agent, or escalates to a human with an automatic ticket. The agent identifies the customer, looks up real orders and account tier from the database, and masks card numbers, SSNs and emails before anything reaches the AI model.

Client
Studio project
Industry
E-commerce & customer support
Category
AI Agents

01 / Challenge

The challenge

Single-prompt support bots struggle once questions span orders, technical issues and billing. They also risk sending sensitive customer data to a third-party model and rarely know when to hand over to a person.

02 / Approach

Our approach

We split support into specialists coordinated by a LangGraph supervisor, put a PII scrubber in front of every model call, and connected the agents to a real customer and orders database. A fallback model provider keeps the service available if the primary model fails.

03 / Solution

The solution

A Python system with a LangGraph supervisor and orders, technical, billing and general specialist agents. It ships with a Streamlit chat interface that shows which agent is active, a FastAPI endpoint for integration, an analytics dashboard, Docker setup and a pytest suite running in GitHub Actions.
Features

What it does

  • 01

    Supervisor routing

    Every message is classified and routed to the right specialist agent.

  • 02

    Human handoff

    Escalates to a person and creates a ticket automatically.

  • 03

    PII masking

    Card numbers, SSNs and emails are masked before reaching the model.

  • 04

    Live customer data

    Looks up real orders and account tier from the database.

  • 05

    Fallback model

    A backup AI provider keeps support running if the primary fails.

  • 06

    API & analytics

    FastAPI endpoint for integration plus an analytics dashboard.

Screenshots

A closer look

Nexus Support chat interface on laptop, tablet and phone
Chat UI showing the active specialist agent and live session details
Diagram: customer, PII scrubber, LLM supervisor, specialist agents
Architecture: PII scrubber → LLM supervisor → specialist agents, exposed via chat UI, REST API and analytics

Technology

Built with

  • Python
  • LangGraph
  • LangChain
  • FastAPI
  • Streamlit
  • SQLite
  • DeepSeek
  • Docker
  • pytest
  • GitHub Actions

See it running for yourself.

Start a project

Have a project like this in mind?

Prefer email? hello@halfclicks.com

Start a project