Jonathan Xu.dev
Lead Software Engineer · Los Angeles, CA

Jonathan
Xu

AI Engineering Platform Engineering Leadership Reliability Python · AWS · Azure

I lead AI engineering and platform with a five-person engineering team across backend, integrations, and BI. I still write the code too. Lately that's meant the AI reporting system that gave Finance two analysts back, the internal AI platform underneath it, and the agentic assistant I'm building on top of a $275M business.

2
Analysts off full-time reporting
5
Engineers led · backend / integrations / BI
Daily
AI production health briefing
280+
Staff onboarded onto new systems
$275M
2026 revenue pace
2.2×
Revenue growth absorbed, same stack
01

About

I lead AI engineering and platform at a consumer products business, and the team of five that works on it. I moved into the lead role in 2026, after three years building the platform I now run.

I run it as a player-coach. I own technical direction and delivery for the team. That means setting project scope, delegating ownership, and signing off on all code review. I've onboarded six engineers onto the platform, I have input on hiring decisions, and I coach the team on getting real leverage out of LLM tooling. I also still write production code every week, which is the part I'd rather not give up.

Most of my recent work is AI in production for people who don't write code. With Finance I built a reporting system that replaced hand-keyed Excel and took two analysts off full-time reporting. Underneath it I built the internal platform, including MCP tooling, model access and routing with per-task cost attribution, and LLM reliability monitoring. Claude Code and Cursor run across the team. I'm building a permission-aware agentic assistant over the ERP now, so stakeholders can ask it questions in plain language.

The rest of the job is the unglamorous half. Sales channels where revenue actually flows, direct EDI with national retailers, reliability on a 6 TB database, and a company-wide cost-accounting transformation that puts manufacturing, finance, and accounting on one playbook. I'm also the one explaining engineering to leadership, through a recurring newsletter to executives and stakeholders, incident communications, and the demos that turn systems work into decisions.

RoleLead Software Engineer · promoted 2026
FocusAI engineering + platform
Team5 engineers · backend / integrations / BI
BaseLos Angeles, CA
StackPython · AWS · Bedrock · Azure
Now shippingAgentic ERP assistant, costing rollout
02

Experience

Lead Software Engineer — Ultra Pro International

Mar 2023 — Present

Lead Software Engineer (2026 – Present) · ERP Application Developer (2023 – 2026) · collectibles & gaming accessories · Los Angeles

Lead a team of five on an ERP platform processing $255M in revenue over the past 12 months and up to 190k wholesale and direct-to-consumer orders a year across Shopify, Amazon, Walmart, Target, Kroger, and GameStop.

  • Own technical direction and delivery for the team. Set project scope, delegate ownership, and sign off on all code review. Onboarded six engineers onto the platform, provide input on hiring decisions, and coach the team on getting real leverage out of LLM tooling. Run the monthly architecture review. Write a recurring engineering newsletter to executives and stakeholders that turns platform work into decisions leadership can act on.
  • Shipped AI-generated financial reporting in partnership with Finance, replacing hand-keyed Excel workflows and moving a non-engineering team onto an AI-native process. Two analysts previously dedicated full-time to reporting now log under 35 hours a month each in the ERP. Planned hiring of additional analysts was shelved, and the standing temp agency contract for three to five data entry analysts was not renewed. Routed to smaller models where the workload allowed, holding per-report cost to a fraction of the naive approach.
  • Built the internal AI platform the rest of the team builds on. MCP tooling connects Jira to Claude / Bedrock for automated ticket triage, deployed to helpdesk and DevOps users, alongside model access and routing with per-task cost attribution. Drove Claude Code and Cursor adoption across the engineering team, and built the enablement layer behind every major rollout, including live training environments used by 280+ staff.
  • Put LLMs into the reliability practice. An automated health check on Lambda, EventBridge, and Bedrock reads platform telemetry every morning and writes up what changed. It catches degradations we'd otherwise meet as incidents.
  • Architecting an agentic assistant layer over the ERP so stakeholders can query orders, inventory, and financials in plain language. The work is tool orchestration, permission-aware guardrails, capability boundaries, and evals. Most AI projects bolt onto a frontier API and let the model do the heavy lifting. I'm betting the other way and building the harness, so open-weight models like Kimi and GLM can carry the workload at a fraction of frontier-API cost.
  • Selected to lead a multi-year, company-wide cost-accounting transformation after prior attempts stalled. One costing standard, with manufacturing, warehousing, finance, and accounting all at the table. Took it to a first executive demo and a greenlit rollout plan.
  • Zero re-platforming while the business more than doubled, from 131k orders / $123M (2023) to 190k orders / $218M (2025), into record back-to-back quarters and a ~$275M pace in 2026, all on the same stack.
  • Own production reliability and incident response on a 6 TB database. Eliminated 25-minute report timeouts, broke a chronic deadlock cycle, root-caused a silent 23-day deploy failure.
  • Built the company's retail EDI platform, ingesting purchase orders directly from Walmart, Kroger, and GameStop, replacing a per-transaction third-party middleman at mid-six-figure annual savings. Own marketplace integrations end to end. Recovered an 8-month Amazon outage, backfilled nearly a year of settlement data, and restored automated nightly reconciliation, unblocking ~$500k/month in revenue.
  • Drove cloud security and infrastructure hardening toward PCI-DSS 4.0 / SOC 2: dual-AZ VPC re-architecture with WAF and SSM-only access, Terraform-managed production changes, enterprise SSO on Microsoft Entra, and a six-environment test fleet restored on demand from production snapshots.

Software Engineer — Roots of Progress Institute

Oct 2020 — Oct 2022
  • Full-stack on Forum Magnum (TypeScript OSS). Grew the Progress Forum from a few users to a few thousand monthly visitors.

Full-Stack Software Engineer — GHC Blue Cross

Jun 2019 — Aug 2020
  • Built database tooling for healthcare information systems on Kubernetes / AWS.
  • Automated HIPAA / HITECH compliance protocols for PHI handling.

Software Engineer Intern — NASA Jet Propulsion Laboratory

Summer 2015
  • VR orbital visualization tooling.
03

Skills & Stack

AI / LLM

Agentic system designLLM harness engineeringModel Context Protocol (MCP)Retrieval over structured dataPermission-aware guardrailsEvalsAmazon BedrockAnthropic Claude APIOpen-weight / self-hosted modelsModel routing & cost attributionClaude Code / Cursor rollout

Leadership

Technical directionProject scopeCode review sign-offOnboardingHiring inputArchitecture reviewsCoaching on LLM toolingExec & stakeholder commsCross-functional programs

Languages

PythonTypeScriptJavaScriptSQLBash

Backend

ERP systems (ERPNext)MariaDBMySQL / AuroraREST APIs

AWS

LambdaEventBridgeBedrockRDS / AuroraS3SSMSecrets ManagerIAMALB + WAFTerraform

Azure

Entra ID / SAML SSOAzure DevOpsMicrosoft GraphTeams APIMicrosoft 365

Infrastructure

KubernetesDockerCI/CDObservability

Integrations

EDI X12Amazon SP-APIShopifyTarget marketplaceSAML 2.0 / SSO

Reliability

Performance tuningLock-contention analysisIncident responseObservability
B.S. Computer Science — UC Riverside 2018

Let's build
something.

Open to roles leading AI engineering teams. The best fits are internal AI platforms and copilots, agentic systems, and the enterprise process work that has to happen alongside them. Happy to talk backend, integrations, and reliability at scale too.