The Linux Foundation Projects
Skip to main content
AIBlogHackathon

Dell Technologies Supercharges SONiC with Gen AI Power: 2024 Hackathon Winner Spotlight

By December 10, 2024No Comments

Authors: Venkatesan Mahalingam, Senthil Kumar Ganesan, Vinoth Kumar Arumugam, Udhaya Chandran Shanmugam and Rathna Sabapathy at Dell Technologies.

The OCP 2024 SONiC Team called on the SONiC community to showcase their most innovative solutions for current challenges. Out of numerous participants, four winners were chosen across different categories. Today, we’re highlighting the Most Innovative Award winner: Ask SONiC AI, a chatbot presented by Dell Technologies.

The Inspiration Behind the Innovation

Artificial Intelligence is revolutionizing every industry, boosting productivity, and opening new possibilities. Inspired by the capabilities of Chat GPT, we wondered if this technology could be harnessed to manage SONiC network fabrics and how impactful it could be.

Imagine simplifying a SONiC developer’s tasks with instant answers to technical and process-related questions.

Envision seamless SONiC user adoption without the need to master specific CLI commands.

Picture troubleshooting time reduced from hours to minutes.

Introducing Ask SONiC AI – Solution Details

Meet SONiC-specific AI agents, designed to assist with all your SONiC queries.

AI Agents: Revolutionizing SONiC Management

AI agents are autonomous intelligent systems designed to perform specific tasks independently, without human intervention. “Ask SONiC AI” is a chatbot solution that leverages these AI agents to dynamically retrieve SONiC information, equipped with a comprehensive knowledge base to respond to user prompts effectively.

  • SONiC Live Agent: Provides real-time answers based on data collected from live switches.
  • RAG Agent: Responds to queries using information from documents.
  • SuzieQ Agent: Answers questions based on telemetry data collected from switches.

Challenges

The primary challenge was creating a prompt that ensures the correct AI agents are selected in the backend for fast response times. Prompt templates and AI agent descriptions are crucial for the LLM reasoning engine to function properly and to avoid hallucinations.

Key Takeaways

This hackathon demonstrated capabilities such as effortless SONiC configuration and debugging through natural language, eliminating the need for deep expertise in SONiC specifics. This approach helps administrators with minimal knowledge understand the SONiC system and manage the SONiC network fabric seamlessly, potentially boosting SONiC adoption.

Next Steps

  • Finetune a smaller model with SONiC-specific data.
  • Utilize the AI agents within the SONiC switch for comprehensive debugging support.
  • Crowdsource SONiC FAQs to create a dataset for training.
  • Open source the finetuned model.