Seller Resolve
An AI support agent that investigates seller issues, applies approved policies, executes verified resolutions, and escalates exceptions.
Built at AgentX Hackathon 2026
Team Break The Bug: Riya, Saksham Upadhyay, Avinash Singh Pal
Built with Python, FastAPI, React, Next.js, TypeScript, PostgreSQL, SQLAlchemy, Alembic, Docker, LLM-based Agent Orchestration
About the project
Online sellers often struggle with repetitive customer support requests involving delayed shipments, damaged products, missing items, returns, and refunds. Handling these issues manually is time-consuming, inconsistent, and difficult to scale, while unrestricted automation can lead to unauthorized refunds and unreliable decisions.
SellerResolve is an agentic AI-powered support and resolution system designed to automate these workflows while keeping sellers in control.
A central AI orchestrator understands customer requests, investigates issues using order, shipment, and supporting evidence records, and coordinates the appropriate resolution workflow. It evaluates cases against seller-configured policies, service areas, refund limits, and authorization rules before initiating any action.
Unlike a conventional chatbot, SellerResolve is designed to close the resolution loop. Backend services enforce policy and authorization constraints, execute permitted actions such as refunds or replacements, and verify their outcomes before communicating success to the customer. Cases involving conflicting evidence, high-risk claims, or actions requiring approval are escalated to the seller with a structured summary and audit trail.
The system follows a hybrid architecture that combines LLM-based reasoning with deterministic backend services, ensuring that AI-driven decisions remain bounded by business rules.
SellerResolve aims to reduce repetitive support workload, improve resolution consistency, and enable online sellers to provide reliable customer support without sacrificing operational control.