Enterprise Knowledge Search · IT Operations Automation · Risk Management
AT&T wanted to put generative AI into everyday employee workflows so software developers, engineers, finance teams, supply chain staff, HR, customer service, and fraud teams could work faster on code, data analysis, support, and risk-related tasks while using a consistent internal architecture and keeping humans in the loop.
Reported impact
AT&T expanded AI access to more than 30,000 employees, accelerated coding and vulnerability remediation, enabled natural-language data analysis, improved Spark pipeline efficiency, and scaled real-time inference for production AI and internal question answering.
AT&T built Ask AT&T as an internal generative AI platform for employees and initially powered it with OpenAI's ChatGPT functionality, while keeping the system interoperable with other large language models including Meta's LLaMA 2 and Falcon transformers. The platform serves users through a conversational interface that accepts plain-English prompts for coding, question answering, and data analysis. AT&T feeds the system with enterprise content such as contracts, financial materials, HR information, AT&T content, software code, and network-related data so it can generate code, answer questions, and analyze operational data by detecting fields, joining tables, and creating code automatically. Beyond the employee-facing tool, AT&T uses NVIDIA RAPIDS AI software to accelerate Spark pipelines, Triton Inference Server to scale production AI/ML and Ask AT&T real-time inference, CuOpt for field technician routing optimization, and Riva for speech-to-text in front of GenAI applications, all aligned to a shared "single source of truth" architecture across common platforms, databases, and code.
AT&T ingests enterprise content and operational data including software code, contracts, financial materials, HR information, AT&T content, and network data into Ask AT&T and related AI workflows. Employees query the system in natural language; the platform uses ChatGPT or alternative LLMs to generate code, answer questions, and analyze data, while RAPIDS accelerates Spark processing and Triton serves real-time inference on shared internal platforms and databases.
"AT&T has built its own generative AI tool for our employees to use. We call it Ask AT&T. It's an intuitive, conversational platform that you can interact with in plain English"
"Ask AT&T was originally built on OpenAI's ChatGPT functionality. But Ask AT&T is also interoperable with other Large Language Models (LLMs), including Meta's LLaMA 2 and the open-source Falcon transformers."
"This tool can detect issues in our software code and, crucially, write and automate patches and fixes as they occur."
"Today, more than 30,000 AT&T employees have access to Ask AT&T"
"We are also testing NVIDIA Riva for tying voice to text in front of several GenAI applications."
"deploying NVIDIA Triton Inference Server for real-time inferencing for use in traditional AI/ML, and Ask AT&T's question and answer chats on AT&T content"
"our initial use case was for software developers to use it to write and refine code"
"we have also applied their RAPIDS AI software for faster/more cost effective (in some cases 2x the speed, 1/2 the cost) Spark data pipelines"
"we'll be able to reduce our response times from days or weeks to literal seconds"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
AWS Sample Apr 17, 2024 Chunghwa Telecom Furthermore, Chunghwa Telecom needed a platform with robust security controls. To improve artificial intelligence (AI) data security and governance, the company migrated its generative AI projects to AWS. The first generative AI solution, the SDLC assistant, has saved developers an estimated 13 percent of their… Amazon Bedrock · Amazon Polly 
In-house Sample Jul 23, 2026 Telefónica Telefónica needs to evolve and maintain a very large API and microservices estate without slowing delivery, while keeping integrations understandable, compliant, secure and maintainable; it also wants to reduce manual… AI agent · generative AI In-house Sample Mar 30, 2026 Telefónica Germany Telefónica's network operations teams had to handle growing network complexity, massive alarm volumes, and incident analysis workflows that made root-cause identification and troubleshooting slower and harder for NOC… Network Operation Agent (NOA) · database · ticketing system 