PERSONAL PROJECT / ENGINEERING CASE STUDY

Commerce AI Hub.

From a commerce request to a reviewed delivery plan.

Published as AI Delivery Hub, this personal project explores structured, human-initiated delivery planning for Adobe Commerce, App Builder and generic software projects. It supports CLI, LangGraph Studio and MCP entry points—not an autonomous production delivery service.

View GitHub repositoryPERSONAL PROJECT / EXPERIMENTAL MVP

Current project status

Experimental MVP; no production deployment or customer-impact claims. Diagram illustrates the workflow, not a complete deployment topology.

01 / AUTOMATE THE PREPARATION. KEEP PEOPLE IN CONTROL.

Automate the preparation. Keep people in control.

Delivery planning needs context, specialist perspectives and explicit review. The workflow classifies a request, selects platform guidance and generates artifacts while preserving review, revision and approval as deliberate steps.

02 / SYSTEM DESIGN

A stateful planning workflow

LangGraph coordinates generation stages using OpenAI. Optional PostgreSQL checkpoints persist graph state, and optional Langfuse tracing exposes execution and generation details. MCP clients can start, inspect, revise, cancel and approve runs.

AI DELIVERY HUB / WORKFLOWREQUEST → PLAN → REVIEW
SYSTEM EXPLORERSELECT A SERVICE ↗

Request → Planning → Generation → Human review

02
LangGraph

Coordinates classification, generation and review stages.

Experimental MVP; no production deployment or customer-impact claims. Diagram illustrates the workflow, not a complete deployment topology.

03 / ENGINEERING DECISIONS

Engineering decisions

01

Human approval gates

Generated artifacts can be approved, revised or rejected before the workflow continues.

02

Platform-aware guidance

A YAML catalog selects profiles for Adobe Commerce, App Builder or generic work, with deterministic fallback when model selection fails.

03

Specialized agent roles

Canonical role profiles accept platform-specific overrides instead of duplicating every role.

04

Model routing by stage

Per-stage model routing allows different generation tasks to use different OpenAI models.

05

Inspectable execution

Optional checkpoint persistence, audit artifacts and tracing support investigation and review.

04 / TECH STACK

Technology

PythonLangGraphOpenAIMCPPostgreSQLLangfuse

Experimental MVP; no production deployment or customer-impact claims. Diagram illustrates the workflow, not a complete deployment topology.

Back to portfolio