Khawaja A.

I build and run the AI systems behind a business, and the people stay in charge of anything that cannot be undone.

A system that does the repetitive part of your operations, drafts what needs a decision, and waits for one of your people to approve anything that sends, posts, pays or deletes. It runs on your hardware and your accounts. You own all of it.

Work I have shipped:

• A health and longevity company: six agents on one knowledge base that audits itself every morning for stale notes, dead links and rules an agent stopped obeying. Live.
• A membership medical practice: an agent fleet, a command centre built for the founder, one knowledge base behind both. Live.
• A US tax and accounting firm: bank statement bookkeeping, financial statements, IRS forms, every output checked by a person before it leaves.
• A university career centre: daily student outreach across dozens of schools, drafted by an agent, sent after a person approves.
• A car dealership software company, a golf tour operator, a home services company. Same pattern each time.

How I work. Context is always the bottleneck, wiring a few openclaw or Claude agents is easy. Every agent session starts from zero, so I write the knowledge down first: what each agent does and does not do, in plain Markdown you can read and edit yourself. Ten to twenty real scenarios per agent, run daily, so a bad edit to the rules gets caught before it reaches a customer. Every change to what the agents know leaves a trail you can roll back. When an agent hits something the rules did not cover, it proposes an edit and a person merges it or does not. Then the framework: OpenClaw, Claude Code, Hermes, LangGraph, n8n or Make, whichever fits. Plain files, so you are not locked in.

I operate, not just build. Most freelancers hand you a bot and disappear. I deploy on your infrastructure with health checks, a watchdog on every rotating token, a second model provider for when the first goes down, two off box backups, and a test suite that runs against the live models every day.

I know what breaks. Config coupling, context overflow, token expiry, rate limits, memory drift, browser session decay, OAuth refresh races. I have hit every one of these in production and the recovery is built in from day one.

How an engagement runs. It starts with an audit: I interview the people who do the work and hand back where the hours go, what to automate, and what not to. Then a build in milestones, each with a capped price and a written acceptance test, paid when you accept it, never a deposit. Then a flat monthly fee to keep it running. Whenever you want it, you take it in house.