MKMichael King Email

Physical operations -> enforceable software

Mechanical engineer deploying AI and enterprise software into industrial operations.

Five years inside a robotics manufacturer. Customer deployments in the US and EU. I build and operate the production systems that carry engineering, production, supply chain, and external builders, and watch them hold up under pressure.

View flagship deployment Resume (PDF) Email

24 idempotent production migrations through a twice-migrated candidate sandbox

Deployment record

Evidence from the field and production.

Customer sites
Commissioned robotic systems in the US and EU; field lessons converted into runbooks
Production system
One governed lifecycle from engineering configuration through installed base, serving four internal functions and external contract builders
Deployment discipline
24 idempotent migrations through a twice-migrated candidate sandbox; zero rollbacks
Live operations
Liveness watchdog built after a silent scheduler failure; first real catch paged in 1 hour 46 minutes
Access model
11 server-enforced roles including row-scoped external suppliers

Flagship deployment

InductOne Tools

A governed ERPNext and Frappe lifecycle for a configurable industrial machine. Designed, built, migrated, administered, and supported in production by one owner.

01 / Problem

A recurring cross-functional scramble.

Product and Engineering manually assembled every builder package. The prior process consumed an estimated 200 hours of constrained senior Product and Engineering work per machine.

02 / System deployed

One governed lifecycle.

Selected configuration becomes controlled engineering state, a bespoke builder package, accepted as-built records, and a durable installed base.

  1. 01Configuration
  2. 02Engineering gates
  3. 03Builder package
  4. 04As-built acceptance
  5. 05Installed base
Artifact ALifecycle and release gates
InductOne governed lifecycle Five lifecycle stages connected by four engineering and acceptance gates. 01Configuration gate 02Engineeringsign-off gate 03Builderpackage gate 04As-builtacceptance gate 05Installedbase
Artifact BPermission boundary
Internal and external permission boundary Internal roles work across governed records while external builders see only assigned machines through a server-enforced boundary. Internal roles Product Engineering Supply chain Operations Server-enforced wall External builders Assigned machines only Row-scoped permissions

03 / My responsibility

Sole architect, developer, and administrator.

I also hold the Operations sign-off on releases. Every requirement is traced to linked evidence before approval. Teams route through that gate willingly: the no is always evidence, never opinion, and resolutions are worked jointly with the owning team.

When testing needed real hardware, I specified and directed the build of a desk-side test rig: two IPCs, an HMI, a PLC, and two PoE cameras networked as a complete cell, so software and communications are exercised on the actual stack before release.

04 / Production rollout

Migration, rehearsal, and scoped access.

Rollout ran through 24 idempotent migrations, a twice-migrated candidate sandbox, persona-based execution harnesses, and external-builder onboarding under row-scoped permissions.

05 / Impact

Manual package assembly became generated output.

estimated 200 hours

removed per machine

modeled $43K-56K

annual recovered capacity at the current run rate

The same engine now produces RFQ packages. Financial impact is a modeled capacity estimate, not audited savings.

Repository evidence: about 33.5K lines, ~50 DocTypes, 40 server-side API methods; sole developer, 5.5 months.

Selected systems

The same operating pattern under different constraints.

OnScript

Production data pipelines, bounded AI, observability

A daily, symmetric measurement instrument for Congressional language, with receipts. It separates deterministic measurement from bounded AI interpretation while publishing its methodology and operational state.

Live siteSource

Agent Guards

Deterministic controls, packaging, distribution surfaces

Deterministic tripwires produce reproducible evidence for known patterns across agent inputs, files, diffs, email, and packages. The system ships as an MCP server, API, and installable plugin; it is a tripwire, not a comprehensive security product.

Source

ForgeSense

Hardware-to-interface systems thinking

Founder of an industrial vibration and edge-monitoring testbed spanning fixture, sensor, embedded compute, analysis, and operator interface. ADXL355, Raspberry Pi, Python analysis, Go edge service, and React with TypeScript.

Private laboratory system

The Exhaust

Research discipline and honest negative results

Public-data collectors and preregistered retrocasts test whether overlooked physical-world exhaust says something useful about the economy. Failed hypotheses remain visible so the work stays falsifiable.

Source

Deployment method

How a deployment runs.

  1. 01Observe the real work. Start with operators, physical process, and handoffs.
  2. 02Model states and invariants. Make configuration, permissions, and lifecycle rules explicit.
  3. 03Encode the guardrails. Put gates, provenance, and failure handling on the server.
  4. 04Instrument the truth. Separate measured evidence, estimates, and interpretation.
  5. 05Ship the whole system. Own migrations, tests, documentation, recovery, and support.

Discovery is where deployments are won: the flagship system exists because I sat with Product, Engineering, supply chain, and external builders, worked directly with the VP of Operations on systemic gaps, and encoded what the floor actually does. Adoption is enforced by design and earned by support: external contract builders work inside the system under scoped permissions, operators run tooling packaged for non-technical users, and the documentation standard I authored is enforced organization-wide.

I run multi-agent AI workflows on business-critical work daily. The reason they can be trusted is not the models. It is the structure around them: orchestration and audit separated from implementation, output reviewed like any production change, claims verified before anything ships. That review has caught embedded PII in a distributable file, staging data presented as production, and access-control claims marked verified that were never tested. Nothing ships on assertion. The evaluation discipline is the same: calibrated holdouts I label blind, frozen evaluation instruments, and precision published next to every claim.

Integration surface: ERPNext to GitLab, production imaging, device log ingestion, and AI agents connected to GitLab and Microsoft 365 for daily work.

Background

Physical engineering and production software.

I work inside industrial robotics and build production systems beyond my formal lane. I am most useful where physical operations, enterprise systems, and AI meet.

Education B.S. Mechanical Engineering, UT San Antonio, summa cum laude, 4.0

Availability

Let’s talk about the deployment.

Open to senior remote roles deploying AI and enterprise systems into physical operations.

View flagship deployment Resume (PDF) Email