J.Crew Group streamlines IT, vendor and customer support with custom AI agents

Solution

To meet its goals, the J.Crew Group and Rackspace teams began by identifying three AI agents to pilot: one for IT help desk support, one for supply chain vendors and one for the customer service center.

J.Crew Group (JCG) Buddy is designed to ease the IT help desk workload by enabling employee self-service. J.Crew Group targeted a 15% reduction in call volume with this agent.

JCIConnect, the vendor agent, supports the supply chain team by answering role-specific vendor questions in multiple languages — saving an estimated two to three hours per day for employees.

JCG Ally, the Contact Center agent helps customer service representatives resolve questions more efficiently by reducing the number of issues escalated to managers.

To bring these bots to life, Rackspace employed a range of technologies across clouds, including LangGraph, Microsoft Azure Entra, the AWS Cloud Development Kit (CDK), AWS Lambda and Google’s Vertex AI API.

“We rely on Rackspace to make the best recommendations in our toolset,” said Gupta. “Each time they made a suggestion, our internal teams validated the recommendation and agreed it was the best approach.”

Rackspace worked closely with J.Crew Group to tailor every element — from authentication to user experience — to fit internal requirements. For example, the team customized prompt engineering with AWS large language models (LLMs) to ensure the AI agents delivered precise, context-aware answers aligned to J.Crew Group’s needs.

To meet multilingual and role-specific demands, Rackspace built the agents on LangGraph, an open-source, multi-agent orchestration framework. “It allows an AI agent to decide whether to answer the question based on internal documents, go to the internet or ask a follow-up question,” explained Subrahmanyam Arunachalam, Data Science Engineer at Rackspace. “LangGraph is one of the most intelligent frameworks available.”

One key challenge was integrating Microsoft Azure Entra for authentication — even though the application was hosted on AWS. Rackspace built a seamless SSO experience that authenticates users through Microsoft Azure®, synchronized logins and applied proper access controls back in the AWS-hosted agent interface.

“The Rackspace team has always been agile,” said Gupta. “They created an on-the-fly solution to help us identify bad documentation. This team has been phenomenal — always going above and beyond.”

To minimize manual upkeep, Rackspace also automated the agent’s knowledge base refresh cycle. Every seven days, it scans for changes and updates automatically, eliminating the need for dedicated document management workflows.

As development progressed, the solution evolved into a multi-cloud architecture. Rackspace used Vertex AI API on Google Cloud to give the AI agent secure internet access, with strict controls to limit source data to J.Crew Group-approved domains such as jcrew.com and wikipedia.com.

The backend architecture leverages AWS Lambda for compute, OpenSearch for vector document storage, Amazon Bedrock for LLM hosting, and Amazon Simple Storage Service (Amazon S3) for document storage. All components were consolidated using AWS Cloud Development Kit (CDK), which streamlined deployment and simplified maintenance.

“The CDK allowed us to create infrastructure templates and reuse them across chatbot projects,” said Arunachalam. “It brought our development timeline down from months to weeks and helped us scale from one bot to three.”