ALLEN FORTUNE
Loc Lemoore, CA Status Open — remote / hybrid Rev 2026.07

AI systems that survive contact with the real world.

Fifteen years in construction — laborer at 15, then project manager and estimator — before I taught myself to build. Now I ship production agentic systems inside real businesses: approval gates, audit trails, paying users, real money. Not demos.

allen@fortune5ai.com See the systems
6 systems in productiondaily use by real teams and customers
3-machine agent fleetorchestrated, coordinated, approval-gated
2 live SaaS productswith paying subscribers
$3M+ public-works bidssupported by AI pipelines I built
Sheet index

Shipped systems

Each one deployed, adopted, and in daily use. Deeper technical walkthroughs available on request.

E-1Larger Electric Inc.

AI estimating pipeline

A six-stage agent chain that works the way a senior estimator works: plan reader, spec reviewer, estimator, takeoff counts, coverage traceability, and an independent adversarial QC pass — producing priced estimates in the company's existing Accubid deliverable format. Includes a quote-swap processor that folds new supplier pricing into an existing revision while preserving labor and recalculating overhead. Adopted because it lands in the formats the team already trusts.

Claude Code subagentsMCPPythonopenpyxl / reportlab
E-2Larger Electric Inc.

Field-service dispatch platformLIVE

The service side of the business ran on phone calls and paper. Now: ticket intake, technician scheduling with day boards, mobile job photos, customer approvals via tokenized e-signature links, branded proposal PDFs, and admin reporting. Row-level security verified end-to-end from a real technician session — not just the admin seat.

ReactSupabase RLSDeno edge functionsStripe / Twilio / Resend
E-3Founder

Allen Grade AssistLIVE

Grading SaaS for teachers, integrated with Canvas LMS, with paying subscribers. Built a custom Canvas MCP server in TypeScript, a Playwright grading pipeline with guardrails (regrade protection, tiered partial credit), and ran the full security pass myself: RLS audit, ownership checks, OAuth cleanup. I teach five college courses — I'm the builder and the customer.

TypeScript MCP serverPlaywrightSupabaseCanvas LMS
E-4Personal infrastructure

Multi-agent executive orchestration fleet

A persistent agent system across three Macs on a Tailscale mesh: Supabase-backed task queue and session coordination, 12+ MCP integrations (7 Gmail accounts, 5 Drive accounts, Playwright, Telegram, custom servers), git-synced persistent memory, and scheduled autonomous jobs that draft but never send. Every outbound action is approval-gated; every write is read back and verified. This is the hard part of agentic AI — trust calibration, failure recovery, and safety architecture for a system that touches real email and real money daily.

Claude Code + Agent SDKSupabaseTailscalelaunchdTelegram Bot API
E-5Founder · getpek.com

Player Evaluation KitLIVE

Sports-coaching evaluation SaaS taken from idea to paying customers: React/Supabase build, Stripe billing, transactional email, trademark screening, and an automated four-platform social content engine (X, Instagram, TikTok, YouTube) with daily drafted posts and engagement targeting — all human-approval-gated.

ReactSupabaseStripeSocial APIs
E-6Selected smaller work

And the rest of the bench

GolfDots, a live social golf app · a Polymarket analysis agent whose edge comes from sharp-odds devig rather than naive LLM probability · an order-intake automation in production for a small food business · branded document pipelines — proposal PDFs, change-order templates, transcript-to-summary workflows, automated lecture decks.

PythonPlaywrightSheets / Drive APIspython-pptx
Materials

Stack

Agents & AI

Claude Code (expert daily driver), Anthropic Agent SDK, MCP server development, multi-agent architecture, context engineering, agent safety patterns. Anthropic Academy, 2026.

Build

Python, TypeScript/JavaScript, React, Supabase (Postgres, RLS, Deno edge functions), REST APIs, Playwright automation.

Integrations

Stripe, Twilio, Resend, Google Workspace APIs, Telegram Bot API, Canvas LMS.

Ops

macOS fleet management, launchd/cron scheduling, Tailscale networking, 1Password CLI secrets, git.

General notes

How I work

Production over demos

Every system above survived contact with real users, real inboxes, real deadlines.

Safety as architecture

Approval gates, write verification, audit trails, row-level security — designed in from the first commit, not bolted on.

Domain-first

I was a working estimator and PM for 15 years before AI. I build inside the business, in the formats the team already uses — adoption is the metric that matters.

I teach

College professor. Explaining complex systems to skeptical non-technical audiences is my day job, not a soft skill.