Infosys has unveiled an innovative solution leveraging NVIDIA’s NIM and NeMo technologies to automate the generation of Topology and Orchestration Specification for Cloud Applications (TOSCA) templates, significantly streamlining the telecom wireless network design process. According to NVIDIA Technical Blog, this development addresses the industry’s need for standardized approaches and reduces the risk of human errors in network design.

Harnessing Generative AI for Network Design

The solution employs generative AI to create a standard utility capable of generating service design templates based on network engineer prompts. This automated tool, powered by NVIDIA NIM, improves the user experience by simplifying parameter edits and enabling real-time processing of user inputs to generate customized YAML templates tailored to specific TOSCA design requirements.

Infosys’ approach integrates pretrained and fine-tuned large language models (LLMs) such as Llama 3-70B and Mistral-7B, delivered as NVIDIA NIM microservices. This integration ensures ease of use for all stakeholders, enhancing productivity by allowing network service designers and OSS solution architects to design carrier-grade networks faster.

Data Collection and Preparation for RAG

Infosys gathered user guide network builder manuals, training documentation, and troubleshooting guides for cloud services to generate accurate, contextual network design responses to user queries. A dedicated chat interface, featuring drag-and-drop functionalities, was created to facilitate easy conversions into the YAML file structure, producing vector embeddings for retrieval augmented generation (RAG).

Technical Challenges and Solutions

To prevent delays, Infosys utilized NVIDIA GPUs to generate vector embeddings swiftly. The solution architecture included a React-based user interface, data configuration management using FAISS for efficient data handling, and robust backend services for user management and configuration. Integration with NVIDIA NIM and NeMo microservices enhanced generative AI learning and inferencing capabilities, ensuring secure authentication and authorization.

Evaluating LLM Performance

Infosys tested various LLM configurations, comparing their performance with and without NVIDIA NIM. The results demonstrated up to 28.5% lower latency and a 15% absolute improvement in accuracy using NVIDIA NIM and NeMo Retriever embedding microservices. This improved model performance enables network service designers to build network designs faster and reduce operational costs.

Sample Use Case

An example use case involves generating a TOSCA template for an Ethernet service with 100 Mbps bandwidth between 1PE and 2CE. The generative AI model responds with a service template design conforming to TOSCA standards in YAML format, showcasing the tool’s capability to produce precise and customizable templates based on user specifications.

Empowering Network Designers

By automating TOSCA template generation, Infosys’ solution addresses the time-consuming nature of manual template creation, enhancing efficiency and consistency for telecom companies. With NVIDIA NIM and NeMo technologies, network service designers can streamline workflows, boost productivity, and ensure uniformity in network design and orchestration.

For more details on deploying generative AI applications, visit the NVIDIA Technical Blog.

Image source: Shutterstock