Position · Portfolio · Methodology

AI doesn't make work faster. It relocates where the hard part is.

Operations management theory applied to AI-era work. A framework for constraint migration, deployed proof across four industries, and a methodology for teaching it.

Applications deployed across property, trading, legal, and AI integration. MCP servers running on Fly.io. A paper applying Goldratt's Theory of Constraints to AI work, with 39 references. Two years of building.

Paul Boucherat · BA (Hons) Business & Operations Management, Nottingham Trent University · Nottingham, UK

The progression

Mar 2024
Started building. OpenAI Assistants API, basic Python. I could design systems on paper but couldn't build them. AI changed that immediately.
Late 2024
Infrastructure and tooling. MCP server deployment, Docker, environment configuration. Steep learning curve, but every broken deployment taught me something I actually needed to know.
Early 2025
First deployed products. Property data API, Land Registry SPARQL integration, a Python package published on PyPI. Within twelve months I could build what I could think of.
Mid 2025
Real-world validation. A 6,000-activity P6 railway schedule needed critical path analysis. Built a 22-tool MCP server that identified 2,024 critical activities and weeks of scheduling delays in minutes. First external deployment on a live operational problem.
Dec 2025
The theoretical connection. The operations management training (Goldratt, Lean, Six Sigma) had been shaping every decision for two years without me fully noticing. The constraint migration framework put words to what I'd been doing.
2026
Live SaaS. Paper published. Framework validated. PineSmith live with Stripe billing. The constraint migration paper published with 39 references. Taught the methodology to a non-technical learner who built 5 MCP servers independently. The method transfers.

The implication

The bottleneck moved. Most people haven't noticed.

AI commoditised execution. The code, the drafts, the data processing: that's the $30/hour work now. The constraint has migrated to orchestration, governance, and judgment. Knowing what to build, how to verify it, when to trust it.

That's not a technology problem. It's an operations problem. Operations management already has the theory for it. Goldratt's Theory of Constraints was built for exactly this. I've applied it to AI-era work in a paper that proposes five new constraints and maps them directly to the TOC diagnostic model.

Every project in this portfolio tested the framework. Each one shaped what came next.

The question is no longer "can AI do this?" It's "where is the bottleneck now?"

Read the full paper →

The Framework

5 Constraints of AI-Enabled Work

When execution gets cheap, these become the real bottlenecks.

Based on Goldratt's Theory of Constraints · Studied under Prof. Roy Stratton · Nottingham Trent University

The proof

An employment law case required forensic analysis of 1,300 evidence items: documents, emails, 850 images. Built a system that processed them into 9 reports with 644 cross-evidence correlations. SHA256-verified chain of custody. Deployed in a live case. Six generations of architecture evolution.

1,300+
items
9
reports
644
correlations
Evidence Toolkit deep dive →

Full-stack algorithmic trading platform. 3 strategy engines, 7 autonomous AI agents, 37 tools, and a custom terminal. Walk-forward tested. Monte Carlo stress-tested across 10,000 simulations.

7
AI agents
37
tools
Trading platform deep dive →

AI-powered Pine Script v6 code generation. The tool didn't exist, so I built it as a product. 50+ component React 19 frontend, FastAPI backend, Stripe billing, Supabase auth. Live on Fly.io.

50+
components
24
endpoints
PineSmith deep dive → Live · Fly.io

Training

Find your own AI working style.

I don't prescribe a method. Everyone's constraints are different, and the right tools depend on how you think. The course starts with your bottlenecks and builds from there.

“The gap between imagination and execution has closed. I have the receipts.”
“He took someone who had never written a line of code and had them building independently within a month.”

5 modules. Two tracks (no-code or builder). 8–10 weeks. You leave with tools you built on your own work, not theory exercises.

See the full course & join the waitlist
1,289 PyPI installs/month
-- requests/30d
-- uptime
-- edge regions
-- apps live

Live data · PyPI + Fly.io · Full dashboard →

Three ways in.

Read the paper
The constraint migration thesis. Free. No signup required.
See the work
Deployed tools across property, trading, legal, and MCP.
Join the cohort
First cohort forming. Your work. Your problems. Your AI working style.

Free Your Main Thread