Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NEMO-4-PAYPAL: Leveraging NVIDIA's Nemo Framework for empowering PayPal's Commerce Agent

Ali Sahami, Sudhanshu Garg, Andrew Wang, Chaitanya Kulkarni, Farhad Farahani, Sean Yun-Shiuan Chuang, Jian Wan, Srinivasan Manoharan, Uma Kona, Nitin Sharma, Linsey Pang, Prakhar Mehrotra, Jessica Clark, Mark Moyou

TL;DR

PayPal's Commerce Agent targets autonomous, conversational shopping but suffers from search-driven latency. By partnering with NVIDIA and applying the NeMo Framework to fine-tune the Nemotron SLM using HyDE-based retrieval and LoRA, the authors significantly reduce retrieval and overall latency while cutting GPU costs, all within a scalable, production-ready multi-agent framework. Extensive SFT and DPO experiments demonstrate favorable speed-quality trade-offs, with substantial improvements in end-to-end performance and maintainable response quality. The work provides a practical blueprint for deploying and optimizing large-scale agentic commerce systems in production environments.

Abstract

We present the development and optimization of PayPal's Commerce Agent, powered by NEMO-4-PAYPAL, a multi-agent system designed to revolutionize agentic commerce on the PayPal platform. Through our strategic partnership with NVIDIA, we leveraged the NeMo Framework for LLM model fine-tuning to enhance agent performance. Specifically, we optimized the Search and Discovery agent by replacing our base model with a fine-tuned Nemotron small language model (SLM). We conducted comprehensive experiments using the llama3.1-nemotron-nano-8B-v1 architecture, training LoRA-based models through systematic hyperparameter sweeps across learning rates, optimizers (Adam, AdamW), cosine annealing schedules, and LoRA ranks. Our contributions include: (1) the first application of NVIDIA's NeMo Framework to commerce-specific agent optimization, (2) LLM powered fine-tuning strategy for retrieval-focused commerce tasks, (3) demonstration of significant improvements in latency and cost while maintaining agent quality, and (4) a scalable framework for multi-agent system optimization in production e-commerce environments. Our results demonstrate that the fine-tuned Nemotron SLM effectively resolves the key performance issue in the retrieval component, which represents over 50\% of total agent response time, while maintaining or enhancing overall system performance.

NEMO-4-PAYPAL: Leveraging NVIDIA's Nemo Framework for empowering PayPal's Commerce Agent

TL;DR

PayPal's Commerce Agent targets autonomous, conversational shopping but suffers from search-driven latency. By partnering with NVIDIA and applying the NeMo Framework to fine-tune the Nemotron SLM using HyDE-based retrieval and LoRA, the authors significantly reduce retrieval and overall latency while cutting GPU costs, all within a scalable, production-ready multi-agent framework. Extensive SFT and DPO experiments demonstrate favorable speed-quality trade-offs, with substantial improvements in end-to-end performance and maintainable response quality. The work provides a practical blueprint for deploying and optimizing large-scale agentic commerce systems in production environments.

Abstract

We present the development and optimization of PayPal's Commerce Agent, powered by NEMO-4-PAYPAL, a multi-agent system designed to revolutionize agentic commerce on the PayPal platform. Through our strategic partnership with NVIDIA, we leveraged the NeMo Framework for LLM model fine-tuning to enhance agent performance. Specifically, we optimized the Search and Discovery agent by replacing our base model with a fine-tuned Nemotron small language model (SLM). We conducted comprehensive experiments using the llama3.1-nemotron-nano-8B-v1 architecture, training LoRA-based models through systematic hyperparameter sweeps across learning rates, optimizers (Adam, AdamW), cosine annealing schedules, and LoRA ranks. Our contributions include: (1) the first application of NVIDIA's NeMo Framework to commerce-specific agent optimization, (2) LLM powered fine-tuning strategy for retrieval-focused commerce tasks, (3) demonstration of significant improvements in latency and cost while maintaining agent quality, and (4) a scalable framework for multi-agent system optimization in production e-commerce environments. Our results demonstrate that the fine-tuned Nemotron SLM effectively resolves the key performance issue in the retrieval component, which represents over 50\% of total agent response time, while maintaining or enhancing overall system performance.
Paper Structure (17 sections, 12 figures)

This paper contains 17 sections, 12 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. E-Commerce Information Retrieval
  4. LLM Agent in Search and Recommendation
  5. NVIDIA’s NeMo Framework
  6. NEMO-4-PAYPAL Framework
  7. LLM-Powered Commerce Search $\&$ Recommendation
  8. Commerce Agent System Architecture
  9. Experimental Results
  10. Data Sources
  11. Evaluation Methodology
  12. Supervised Fine Tuning (SFT)
  13. Direct Preference Optimization (DPO)
  14. Inferencing Latency
  15. SDK Framework Comparison
  16. ...and 2 more sections

Figures (12)

  • Figure 1: PayPal is transforming from a payments company to a commerce platform with, our commerce Paypal Agent, which leverages NVIDIA’s Nemo.
  • Figure 2: Commerce Search and Recommendation System at PayPal. The system comprises four key components: (1) Query Understanding, Expansion, and Formulation (LLM); (2) Retrieval; (3) Ranking; and (4) LLM Evaluator.
  • Figure 3: NeMo Framework nvidia_nemo_framework_2024 NVIDIA NeMo is an end-to-end platform for developing custom generative AI, anywhere. It includes tools for training, fine-tuning, retrieval-augmented generation, guardrailing, data curation, and pretrained models. It has offerings across the tech stack, from frameworks to high-level API endpoints.
  • Figure 4: Agentic Commerce AI system: (1) Commerce Platform (2) Agent Orchestration Framework, (3) LLMs Integration (4) Personalization Engine.
  • Figure 5: Example: Query formulation output by LLM and is used for data generation for fine-tuning
  • ...and 7 more figures