Business-systems engineer
I build systems businesses run on.
For two decades, I have built and operated production software for ecommerce: pricing, purchasing, marketplaces, inventory, accounting, advertising, fulfillment, customer operations, and the infrastructure that keeps all of it working. Today I also build AI agents - but on top of real systems, with evaluated decisions, deterministic boundaries, and production safeguards.
- 20 years building for a live ecommerce business
- ~15 years of continuous production for the custom storefront
- Amazon, eBay, Walmart, Newegg, payment, carrier, and accounting integrations
- Financial, inventory, advertising, supply-chain, fulfillment, and AI systems
- Public work in Python, Rust, TypeScript, and agent infrastructure
Software shaped by operating experience
The hardest requirements rarely appear in an API document. They emerge when a settlement is late, a product exists under five identifiers, a worker dies after an external side effect, inventory is between China and a warehouse, or a carrier returns a label that the printer cannot use.
I have spent years on both sides of those problems: running the operation and building the software. That changes what I optimize for. I care about auditability, idempotency, recovery, reconciliation, operator visibility, and safe migration - not just the happy path.
What I build
Business systems
I model operational decisions and financial rules, then carry them through data pipelines, interfaces, background jobs, and integrations.
key systems - bookkeeping & inventory accounting, purchasing allocation, marketplace operations, fulfillment, law-firm scheduling
Production integration
I am comfortable entering a live environment with history, old schemas, external APIs, manual exceptions, and software that cannot be allowed to stop.
key systems - the long-lived Django storefront, Xero and marketplace synchronization, multi-carrier shipping layer
Applied AI & agents
I build agents for bounded work: pricing judgment, research, classification, self-healing browser automation, and development. Deterministic systems keep identity, money, state transitions, and side effects.
key systems - agentic pricing, AgentLib, Code Agent, customer email triage, self-healing browser automation
Reliability & evidence
I design for delayed data and partial failure: append-only ledgers, event logs, durable claims, desired state, frontiers, reconciliation, monitoring, and recovery.
key systems - Report Archive, advertising control loops, health monitoring, Envoy, event-sourced audit logs
Open repositories
Selected open-source repositories reflecting the same engineering discipline as my private production systems.
agentlib
PythonA lightweight framework for building agents with typed tools, Pydantic models, and provider-neutral transports. Powers the production pricing and classification agents.
code-agent
PythonAn interactive coding agent in the spirit of Claude Code or Codex, built around a Python REPL: sessions that resume and fork, and context the agent manages itself across thousands of turns.
envoy
Python / PWAA web-first terminal emulator and hosted PWA built for server-side coding across devices: multi-tab PTY sessions, touch gestures, remote path & image previews, drag-and-drop file upload, and an integrated voice/text agent with screen awareness.
pixeldeck
Python / RustLocal batch image post-processing and touch-up studio for product photos: auto-crop, background removal, upscaling, inpainting, and tone and color correction, running on local Rust engines. The studio behind the catalog imagery pipeline.
More on GitHub
@jacobspartsProtocol gateways, image processing in Rust, browser automation, and other public utilities.
How I work
Understand the operation
Start with the people, decisions, exceptions, and business consequences. Reconstruct rules from actual behavior when documentation is incomplete.
Identify hard boundaries
Separate source-of-truth state from derived views and agent context. State what must never happen: duplicate charges, invented inventory, unbalanced entries, lost appointments, or uncontrolled external actions.
Ship alongside reality
Integrate incrementally with live systems. Use compatibility layers, dual-run checks, idempotent jobs, reconciliation, and reversible deployment rather than demanding an all-at-once rewrite.
Instrument and improve
Store evidence, expose status, evaluate outcomes, and make failures repairable. A production system is a feedback loop, not a launch event.
What I am looking for
I am looking for a steady engineering role where I can build and maintain software that matters to the business, understand the operation deeply, and keep building - established companies and startups in ecommerce, operations, fintech, supply chain, applied AI, internal platforms, and developer tools. I also consider focused contract work where the problem benefits from this combination of technical and operating experience.