Embedded Files
Shared Production AI-Agent Platform
Shared Production AI-Agent Platform
Internal Walmart Business platform built to provide reusable foundations for developing and operating production AI agents across multiple business units.
Python • FastAPI • LangGraph • Azure OpenAI • MCP • MySQL
Impact at a Glance
Impact at a Glance
4 Business Units
4 Business Units
Platform Adoption
Shared framework adopted across four Walmart business engineering teams.
Platform + Agents
Platform + Agents
End-to-End Scope
Built both reusable platform foundations and production agents running on top of them.
Reusable Architecture
Reusable Architecture
Developer Enablement
Created common agent abstractions so teams could reuse orchestration and integration patterns instead of rebuilding them.
Production
Production
Operational Scope
Designed as an internal production platform rather than a prototype or academic AI project.
Problem / Context
Problem / Context
Why We Built It:
Why We Built It:
Multiple Walmart Business teams were developing AI-agent capabilities and repeatedly needed similar infrastructure for agent orchestration, API integration, memory handling, and tool connectivity.
b2b-ai-core was created as a shared internal platform so teams could build business-specific agents on top of reusable foundations instead of independently recreating the same integration scaffolding.
How the Platform Works
How the Platform Works
1. API Layer
1. API Layer
FastAPI exposes the application and agent-facing service interfaces.
2. Agent Orchestration
2. Agent Orchestration
LangGraph provides the workflow/orchestration layer for ReAct-style agent execution.
3. LLM Integration
3. LLM Integration
Azure OpenAI provides the underlying LLM capabilities used by agents.
4. Tools & State
4. Tools & State
MCP integrations, reusable abstractions, and swappable memory backends allow agents to interact with tools and maintain workflow context.
Business-Specific Agents
Business-Specific Agents
Individual teams can build production agents on top of the shared framework while keeping their business logic separate from the common platform layer.
What I Personally Owned
What I Personally Owned
Architected and implemented core platform foundations for the shared internal AI-agent framework.
Built reusable BaseAgent-style abstractions to standardize common agent behavior.
Designed swappable memory abstractions so agent workflows were not tightly coupled to one memory implementation.
Implemented agent orchestration using LangGraph and ReAct-style workflows.
Worked with Azure OpenAI as the platform's LLM provider.
Built MCP-based tool integrations for agent-to-tool interaction.
Contributed directly to the FastAPI service/API layer.
Built production agents on top of the framework, not only the underlying platform.
Helped productionize the framework for adoption across 4 business units.
Engineering Decisions I Drove
Engineering Decisions I Drove
Shared Platform vs. Duplicate Agent Stacks
Shared Platform vs. Duplicate Agent Stacks
I helped design the platform around reusable common foundations rather than allowing each business unit to independently rebuild agent orchestration and integration infrastructure.
Reusable Agent Abstractions
Reusable Agent Abstractions
I designed reusable agent abstractions so business-specific agents could share common orchestration and tool-integration patterns while keeping their domain logic isolated.
Swappable Memory
Swappable Memory
I helped structure memory behind reusable interfaces so agent implementations were not tightly coupled to a single memory backend and could evolve without rewriting core agent logic.
MCP-Based Tool Integration
MCP-Based Tool Integration
I incorporated Model Context Protocol (MCP) into the platform's tool-integration approach to create a cleaner separation between agent reasoning and external capabilities.
Platform Architecture
Platform Architecture
Business-specific logic remained separate from the shared platform layer, allowing teams to reuse core agent infrastructure while implementing their own workflows.
Production Adoption
Production Adoption
From Framework to Production
Shared Foundation
Shared Foundation
Developed as a common internal AI-agent framework rather than a single-purpose application.
Production Agents
Production Agents
Built actual production agents directly on top of the shared platform.
4 Business Units
4 Business Units
The platform was adopted across four Walmart business units.
Bottom CTA
Bottom CTA
Explore another engineering project or ask my AI portfolio detailed questions about my architecture decisions, technical ownership, production experience, or fit for a specific role.
Page updated
Google Sites
Report abuse