Open to remote roles

Portrait of Patrick McClain II, smiling, printed as an off-white dither on a warm gold plate. Portrait of Patrick McClain II, smiling, printed as an ink dither.

Patrick McClain II

Senior Software Engineer Houston, TX · Remote

I’ve spent 7+ years modernizing production web and cloud systems, most recently the safety-critical pharmacy platform at H-E-B.

Currently building SPARC: an evidence-driven software factory where AI agents pin down intent, iterate quickly, and ship proven code.

Resume (PDF) Email GitHub

GitHub activity

9,590 contributions in the last year

Mostly SPARC, built on nights and weekends.

The day-by-day graph needs JavaScript. The totals beside it are the real numbers.

Active days
294
Longest streak
39 days
Best day
375 · Oct 18, 2026
github.com/patrickmcclain
Contributions per month, Oct 26, 2025 to Oct 28, 2026
Total9,590

At work now

SPARC

an evidence-driven software factory where AI agents pin down intent, iterate quickly, and ship proven code.

Personal project · Built solo · Jun 2025 – present

A software factory, built solo on my own time and hardware. You give it a request, anywhere from a one-line fix to a large multi-story feature. AI coding agents (Claude Code and Codex) plan it, build it, review it and test it until it merges.

No agent gets to say its own work is done. A small deterministic program checks evidence the builder couldn't control, and only then does the work count.

SPARC's 3D monitor: an isometric factory where each AI agent is a worker at its station, lit while it works.
Fig. 1 SPARC's 3D monitor: each AI agent is a worker at its station, lit while it works.

How it works

  1. Step 1

    Intent

    The request becomes an approved statement of purpose, scenarios and quality bars. Only the human changes it.

  2. Step 2

    Plan only the next proven step

    A lead agent plans the work as small stories and details only the next batch, so a wrong assumption costs a step, not the whole plan.

  3. Step 3

    Build and check, independently

    Builders write the code. Separate QA and reviewer agents write checks the builder can't edit, and a check counts only if it fails on a deliberately broken version.

  4. Step 4

    Merge on evidence

    A pure-function kernel decides what may happen next. The merge queue lands only the exact commit that was tested, so no merge adds a failure.

  5. Step 5

    The human accepts

    The human accepts the finished result.

Results

Measured on real runs, Oct 2026.

  • Memory used by the executor

    10× less memory

    1.9 GB to ~170 MB

  • CPU time per busy pass

    40× less CPU

    9.8 s to 0.2 s

  • Time the database stays locked (median)

    ~2,500× faster

    ~50 s to ~20 ms

All while the event history grew more than 2×.

Said “saved”
64
Actually saved
61
Lost
3

Caught before it shipped

The test: a small notes app. I asked SPARC to add a server that saves your notes, and gave the same request to one AI agent working alone.

It said “saved” 64 times. Only 61 notes were.

Every normal test passed. Then SPARC’s own QA agent, an AI tester, tried to break it by starting two copies of the server at the same moment. They overwrote each other’s saves, and 3 notes the app had confirmed were gone.

Bugs like this pass normal tests and only show up when two things happen at once, which is how real users lose data.

  • One AI agent alone: shipped with this bug.
  • SPARC: blocked it, fixed it, then broke its own fix 3 times until the 4th held.

SPARC test project, run 2 of 17 · Oct 8, 2026 · counts measured before the fix

Proven models

18

TLA+ models are tied to the running code by trace replay. Each model had to catch a planted bug before it counted.

Architecture

Before: a chain of AI managers passing notes. After: one shared record and a small program that decides.

Watch the same small request go through each one.

BeforeAn organization of agents
Agents started
16
Hand-offs between agents
44
Times it needed you
4

Done: 16 sessions, 44 relays between agents, and the Human was needed 4 times. Building it was one stop on the way.

AfterA record and a program
Agents started
4
Hand-offs between agents
0
Times it needed you
1

Done: 4 sessions, no relays, and one touch for the Human. The rest was the record and a small program deciding.

hand-offs between agents
44to0
agents started
16to4
times it needed you
4×to1×

Why it matters: fewer hand-offs means fewer places for work to get lost or stall, and every decision is recorded and can be replayed.

Under the hood: SPARC replaced a hierarchy of LLM coordinator agents with a pure-function kernel and a crash-safe executor over an append-only SQLite event log, so every decision is deterministic and auditable.

Sketch of one small request. Measured in the old system: 17 sessions minimum for one story, with coordinators using about 58% of agent time in a small evaluation.

Close-up of SPARC's 3D monitor: an isometric factory where each AI agent is a worker at its station, lit while it works.
Fig. 2 The MAIN warehouse, with lamps lit where agents are at work.

The 3D monitor

A real-time isometric 3D view of the factory, built in three.js and WebGL2.

Every agent, story and merge appears as a working factory, with replay to any moment in the event log.

The plant at the top of this page is a live sketch of it.

On shift since 2019

Experience

  1. Apr 2022present

    H-E-B Senior Software Engineer

    Houston, TX (hybrid) On shift

    RxEVO is H-E-B's in-house pharmacy platform. It runs intake, clinical safety checks, insurance billing, pickup and delivery for 300+ pharmacies. The team is modernizing its 20-year-old Java core into Angular and GCP services.

    • Building H-E-B's new cloud Central Fill Service to route prescriptions to off-site automated fulfillment centers. Stood up its GCP/GKE infrastructure in two days and designed its vendor-neutral API.
    • Built Red Flags end to end as its lead developer (Angular, Swing, Spring Boot). It's a chain-wide alert system for suspicious patients and prescribers that guards H-E-B against drug diversion and DEA compliance risk.
    • Core engineer on replacing the Swing desktop client with Angular: new screens were embedded panel by panel and rolled out store by store behind feature flags, so any change could be rolled back.
    • Rebuilt the pickup and counseling workflows every H-E-B pharmacist uses at the counter. Added per-vial scan verification to stop wrong-prescription handoffs, and Batch Counsel so pharmacists counsel a whole visit at once.
    • Strengthened H-E-B's disaster readiness: helped build RxEVO Lite, the backup app pharmacies switch to if the core system goes down, and wired it into on-call monitoring so it's ready when needed.
    • Secured the health data behind customers' pickup texts and emails by adding Vault transit encryption to each on-prem sender and decryption to the cloud notifier.
  2. May 2019Apr 2022

    Frost Bank Software Engineer II

    San Antonio

    Promoted from Software Engineer I in Sep 2021.

    • Led Frost Online's migration from AngularJS to Angular as its main developer. Introduced TypeScript and wrote the conversion guide the other engineers followed.
    • Merged two critical back-office tools into one Angular app in a year-plus, multi-team effort, and developed the web front end for a critical external app integration.
  3. Jan 2019May 2019

    Tata Consultancy Services Software Engineer

    San Antonio

The short version

Why hire me

I think the next decade of software will be written mostly by agents and directed by people. Human work moves to the ends: defining what we want and verifying the results, then iterating quickly. It's the most exciting shift of my career, and I don't want to watch it from the sidelines. I've spent seven years modernizing systems people depend on. Now I want to help build what comes next.

Patrick McClain II

Inventory

Skills

Languages
  • TypeScript
  • JavaScript
  • Java
  • Rust
  • SQL
  • Python
Front end
  • Angular
  • React
  • three.js and WebGL
  • HTML
  • CSS and Sass
  • Design systems
Back end
  • Node.js
  • Spring Boot
  • GraphQL
  • PostgreSQL
  • GCP
  • GKE and Kubernetes
  • Terraform
  • Vault
  • Pub/Sub
Delivery
  • GitLab CI
  • Feature flags (Unleash)
  • Datadog
  • PagerDuty
  • Locust load testing
Agentic engineering
  • SPARC
  • Claude Code
  • Codex
  • Copilot CLI
  • Multi-agent orchestration
  • MCP
  • Event sourcing
  • TLA+

Waiting on you

Open to remote senior roles.

Happy to travel for team gatherings.

p.mcclain281@gmail.com

Houston, TX · RemoteCentral Time