← Back to Home
Case Study

Enterprise Multi-Agent RAG Platform

AI Tech Lead

Client
Leading multinational digital marketing company
Role
AI Tech Lead
Scope
Department-level AI platform – team building, architecture, delivery to production
Enterprise Multi-Agent RAG Platform architecture diagram
Multi-agent RAG platform: consumers reach an intelligent orchestrator through an API gateway or an MCP server, which routes each query to specialized agents backed by retrieval, LLMs, and durable workflows, all under a shared governance layer.

Overview

Within a leading multinational digital marketing company, I led the creation of an AI capability for a key business department from the ground up: hiring and growing a dedicated AI engineering team focused on agentic solutions, defining the platform architecture, and taking multiple AI products from concept to production.

The result is a multi-agent platform of specialized RAG agents, unified by an intelligent orchestrator that routes each user query to the right agent. Our team was also the first in the company to integrate with the enterprise agent network platform, exposing our agents as MCP tools to external consumers and setting the precedent for how other teams could follow.

What I Built

Intelligent orchestration layer

I designed and implemented an orchestrator that interprets user intent and seamlessly routes queries to the most appropriate agentic solution, giving users a single entry point to a growing set of specialized agents.

Multiple production RAG solutions

I built several domain-specific RAG pipelines combining vector search, knowledge retrieval, and conversational AI, powered by frontier LLMs and designed for accuracy, scalability, and grounded responses.

Autonomous multi-agent workflows

I developed agents capable of task planning, tool calling, structured output generation, and multi-step workflow execution, with human-in-the-loop checkpoints for high-stakes business operations. Durable, long-running workflows are managed with Temporal.

MCP interoperability with external platforms

First in company

Working closely with external partners and providers, I led the company's first integration with an enterprise agent network platform. The integration exposes the platform's capabilities as MCP tools through FastMCP, so external consumers can integrate our agents into their own ecosystems. This pioneering work established the reference approach for agent interoperability across the organization. The agents also consume external platforms and services as tools.

Production-grade reliability

I defined the guardrails, AI safety policies, evaluation frameworks, and benchmarking strategy used to validate agents before and after release. I also established end-to-end observability with Langfuse and Sentry to monitor quality, cost, latency, and user experience.

Leadership

  • Built the department's AI team from scratch, defining roles, running hiring, and establishing engineering practices for agentic development.
  • Owned the full lifecycle, from solution design and architecture through implementation and production deployment.
  • Acted as the bridge between product, engineering, business stakeholders, and external technology partners, translating business needs into AI-driven solutions.

Impact

  • First team in the company to integrate with the enterprise agent network platform, creating a blueprint for agent interoperability that other departments can reuse.
  • Established a reusable foundation for agentic AI within the department, allowing new agents to be added and exposed to partners without rebuilding infrastructure.

Tech Stack

Backend & Agents
PythonFastAPIFastMCPTemporal
Retrieval
Milvus (vector database)
Models
Frontier LLMs (OpenAI)
Frontend
Angular
Observability
LangfuseSentry
Integrations
MCPEnterprise agent network platform

Core Competencies

Multi-agent systemsAgentic workflowsRAGQuery routing & orchestrationTool/function callingStructured outputsHuman-in-the-loopGuardrails & AI safetyLLM & agent evaluationAgent observability

Back to all projects, see my experience, or get in touch to talk about building something similar.