CUBANHACKER
Client Work

Built. Shipped. Still running.

Systems I built for clients that are still running, with the numbers to show for it — and the book that teaches the pattern behind them. Client names withheld; every number is real. For my own products and research, see Bootstrapped.

Case studies on karel.build
4

Production AI systems

500+

Cases through wcpdr

14 mo

Platform running non-stop

286

Pages, published

01Case Files

Four systems, in production

Client identities withheld. Numbers straight from production logs and live databases.

~/work/01-legal-ai-platform
Running · daily
01Agentic platform · Legal

A law firm's new department, shipped as software

California workers'-comp firm · multi-location · name withheld

problem
Case processing was gated by people: medical-legal reports read by hand, case knowledge stuck in a few heads.
build
An end-to-end agentic pipeline. It ingests case documents, runs multi-step LLM analysis against the firm's own data with vector retrieval, extracts the findings (examining physician, injury date, impairment rating, P&S status), writes the memo, and routes it to the right staffer. Every run is cost-instrumented.
result
Live since 2025 and still running daily — hundreds of case documents processed firm-wide. A comparable build commissioned today runs well into six figures. I built it solo.
n8nOpenAI + AnthropicPinecone RAGMicrosoft 365MySQLAzure
14 mo

In production, non-stop

$150k+

Replacement cost (market est.)

1

Builder, end to end

~/work/02-wcpdr
Running · daily
02Product · Novel AI tool

wcpdr — an AI that rates permanent disability

Productized · live at wcpdr.com

problem
California workers'-comp permanent-disability rating is slow, error-prone expert work — and nothing off the shelf does it.
build
I modeled the rating rules, wrapped them in agentic logic over the medical report, and shipped it as a real application with real firm users. A product, not a notebook.
result
In daily production use since September 2025: 500+ cases processed for its client firm, about 40 a month, at under $3 of AI per case.
TypeScriptLLM agentsPostgresVercelwcpdr.comProduct & changelog
500+

Cases processed

~40

Cases a month

$2.75

Avg AI cost per case

~/work/03-teams-assistant
Running · daily
03Assistant · Firm-wide

An AI the whole firm asks — inside Teams

Microsoft Teams · every staff member

problem
Answers about live cases lived with a handful of people, and everyone else waited on them.
build
A retrieval-grounded assistant over the firm's own case corpus, wired into the Teams workflow staff already live in. No new app to learn — the answer comes to them.
result
In daily firm-wide use, a year on. Built in 2025, when getting LLMs to work reliably on real case data — not a demo — was genuinely hard.
Microsoft TeamsRAG / vector searchOpenAI + AnthropicMicrosoft 365
Daily

Firm-wide use

100%

Of staff can use it

24/7

Real-time, on demand

~/work/04-intake-agent
Delivered
04Agent · Intake automation

An intake agent that opens the case file for you

From a message to a finished case package

problem
New-case intake is a slow, people-heavy funnel.
build
Multi-step agents extract and structure the intake, assemble the package against the firm's templates, and hand it to staff for the human step — governed, traceable, and costed per run.
result
60 real case packages delivered in production at about 13% of the equivalent staff cost — an ~87% cut on that workflow.
PythonAgent orchestrationMCPCost accounting
60

Case packages delivered

13%

Of equivalent staff cost

87%

Cost reduction

02Published

The book behind the builds

Retrieval-augmented agents, written down before they ran a law firm.

LangChain for RAG Beginners

Build Your First Powerful AI GPT Agent

A 286-page, hands-on guide to building AI agents grounded in your own data with retrieval-augmented generation. It's the same pattern running inside the law-firm platform and the Teams assistant above, taught step by step.

Format
Paperback
Pages
286
Published
Aug 14, 2024
Publisher
Independently published
ISBN-13
979-8332280139
Language
English
Get it on Amazon
03Track record

Twenty years, one commit at a time

From keeping other people's systems alive to building the ones that run themselves.

~/karel — git log

$ git log --graph --reverse --format="%ad %s"

  1. 2004init: found Zoltis Technologies — managed IT for Los Angeles law firms and mid-market companies
  2. ····chore: industrial engineering + business training — map the process first, then price the waste in it
  3. 2023feat: go deep on AI engineering — start teaching AI and automation as Cuban Hacker
  4. 2024docs: publish LangChain for RAG Beginners — 286 pages on agents grounded in your own data
  5. 2025feat: ship clases.community — a community platform with a full agent runtime
  6. 2025feat: build a law firm's agentic AI platform — still running in production
  7. 2025feat: launch wcpdr — AI permanent-disability rating, in daily use since September
  8. 2025feat: put a case-aware AI assistant in front of the whole firm, inside Teams
  9. 2026perf: intake agent delivers 60 case packages at ~13% of the staff cost
  10. 2026feat: ship Cuadrila — governed AI work crews with permission ladders and audit receipts
  11. 2026release: open karel.build — fixed-scope AI systems for mid-market firmsHEAD → main
Next Step

Want a system like these running in your business?

I build them through karel.build — fixed-scope AI systems for mid-market firms. Or follow along on the channel.

Work with meGet the newsletter