Introduction

The landscape of recommendation systems has traditionally been dominated by algorithms leveraging historical transaction data and user interaction patterns. These systems analyze past behaviors to predict future preferences, offering product suggestions based on historical trends and in-session search queries. While effective in many scenarios, these traditional methods need to be revised regarding discretionary items, where user decision-making is more fluid and influenced by discovery and new information. Conversational AI presents a promising alternative for such cases, enabling a more dynamic and interactive user experience. This talk will explore how large language models (LLMs) can be harnessed to build conversational AI that enhances recommendation systems, providing real-time, context-aware suggestions that improve the buying experience.

Overview of the Session

1. System Architecture: Compare traditional recommendation systems, which rely on historical data, with conversational AI-based systems that adapt to real-time user interactions for more relevant suggestions.
2. Domain Adaptation of LLMs: Explore domain adaptation techniques for LLMs, including prompting, instruction fine-tuning, and Low-Rank Adaptation (LoRA), with code demonstrations for specific tasks.
3. Model Serving: Discuss deployment strategies for conversational AI, covering on-premises and cloud-based solutions, with practical examples and code snippets.
4. Open Source Model Demonstrations: Showcase real-world applications using models like Llama and FlanT5, with step-by-step setup, customization, and deployment guidance.