Public work example
An AI delivery orchestrator built for governed, shareable workflows
The ai-delivery-orchestrator is a governed workflow service that uses LangGraph, GitHub automation, and AWS to sequence issues, coordinate AI builds, and review pull requests while preserving human approval boundaries.
It is licensed under Apache 2.0 so it can be shared, and the intended use is a client-owned fork: a versioned release a client can run in its own GitHub organization and AWS account. It is still a work in progress, not a finished product or a claim of a particular result.
What it does
Sequence work and keep humans in control.
The orchestrator coordinates the steps between a structured request and a reviewed pull request: it sequences issues, coordinates AI builds, reviews the results, and records where a person must approve, reject, or take over.
Approval boundaries are part of the design. Model execution does not receive publishing credentials, and automatic merge is treated as a separately gated future capability rather than a default.
How it is meant to be shared
A client-owned foundation, not a hosted service.
The repository is licensed under Apache 2.0 and intended for sharing. The documented direction is a versioned release a client can fork into its own GitHub organization and provision in an AWS account it controls.
A client fork keeps ownership of infrastructure, Terraform state, credentials, policies, repositories, generated code, logs, backups, budgets, and final publication decisions. A hosted control plane or consultancy-operated GitHub App is a separate future product.
Current status
A work in progress, evaluated honestly.
The repository is in its foundation phase. It currently provides a validated TypeScript worker, versioned domain and state-machine contracts, PostgreSQL persistence, a stub-only runtime with durable checkpoints, container build, local runtime, tests, CI, and a reviewed Terraform foundation.
It does not yet connect to GitHub, OpenAI, or a deployed AWS environment and cannot mutate another repository. It is published early to make progress inspectable, not to present a finished product or guarantee a particular result.
Have a workflow worth exploring?
Bring one real process, and we can discuss whether AI is a reasonable fit and what a sensible first step could be.