We developed an AI-driven multi-agent system for an e-commerce platform aimed at improving the smartphone buying process for users who wish to upgrade or sell their existing phone. The system helps users receive personalized phone recommendations based on their preferences, trade-in value estimation, and other essential factors like camera quality, screen size, and price. This system provides a conversational and efficient experience for users, helping them make informed purchasing decisions while seamlessly integrating trade-in assessments and product recommendations.
Complex Decision-Making: Users face difficulty navigating a wide variety of phone models, making it challenging to choose the best one for their needs. Inconsistent Trade-In Value: Providing a fair and transparent trade-in value for users' existing phones was a major hurdle, as manual methods were prone to errors and inconsistencies. Lack of Personalization: Traditional phone recommendation systems didn’t fully account for the nuances of users’ preferences, like camera quality, screen size, or specific features. User Engagement: Ensuring that users remain engaged throughout the phone selection process while making the experience interactive and conversational.
The multi-agent system developed incorporates innovative features such as Self-Query Retriever and Pinecone vector database for similarity-based searches. The integration of these technologies allows for: Real-time, context-aware trade-in value assessments based on the user's existing phone's condition. A Self-Query Retriever that autonomously generates metadata queries, improving search accuracy by retrieving the best-matching phones from the Pinecone vector database. Dynamic, personalized responses delivered via OpenAI GPT-4, ensuring that the user receives highly relevant phone suggestions in a conversational manner. This system pushes the boundaries of traditional e-commerce recommendation engines by combining multi-agent workflows with advanced AI technologies.
User Interaction & Data Collection: The Ask Entity Agent engages users by asking questions about their current phone’s model and condition to determine the trade-in value. It also gathers preferences related to the new phone, such as camera quality, screen size, budget, and brand preferences. Trade-In Value Estimation: The Trade-In Value Agent processes the user’s input and calculates the estimated trade-in value based on the condition and model of the existing phone. This value is presented to the user to inform their budget for purchasing a new phone. Phone Recommendations: The Recommendation Agent uses the metadata provided by the Ask Entity Agent (including trade-in details) and runs a similarity search using the Self-Query Retriever with the Pinecone vector database. It returns the most relevant phone recommendations, which are then shared with the user via OpenAI GPT-4. Seamless Workflow Orchestration: The Supervisor Agent coordinates all the interactions between agents, ensuring a smooth and efficient process that delivers relevant results to the user in a timely manner.
The client, an e-commerce platform specializing in electronics, collaborated closely with us to define the ideal phone attributes and user preferences to capture in the recommendation process. Frequent feedback sessions helped refine the user experience, ensuring that the system’s recommendations were aligned with customer expectations and business goals. Our team worked alongside their data scientists to ensure that the trade-in value algorithm was accurate, transparent, and fair, and that the phone inventory database was robust and comprehensive.
“AI-powered multi-agent systems are revolutionizing e-commerce by delivering personalized recommendations and seamless trade-in experiences, making smartphone purchases smarter and more efficient.” — Alex Peterson, Product Manager