AI Agentic Clienteling Platform

Multi-agent AI platform automating retail clienteling workflows

Salesfloor

Python 3.12 LangGraph LangChain MCP A2A LangSmith OpenTelemetry

Overview

Architected a modular multi-agent AI platform to automate retail clienteling workflows — generating personalized associate tasks and communications end-to-end, from customer data qualification through content drafting. Designed a planner-led agent architecture using A2A (Agent-to-Agent) protocol and MCP (Model Context Protocol), with a central planner orchestrating specialized downstream agents via standardized JSON-RPC with discoverable agent cards.

Highlights

Planner-led agent architecture with A2A protocol and MCP for standardized agent communication
Deterministic SQL and rules-based logic for filtering; LLMs reserved for reasoning-heavy tasks
Full observability with LangSmith for LLM tracing and OpenTelemetry for distributed telemetry
End-to-end debugging and performance monitoring across agent hops
Reusable architectural pattern with roadmap for production hardening and AI governance