About · he/him
One name on every project.
I am Salman Lakhani, an AI founder and builder in Tampa Bay. I run LLCreativity LLC, a studio kept small on purpose, and I build products inside it: WealthRx AI for CPAs, KreoGrid AI for creators, and TZUR Impact for the people who have to know what a cloud change does to a federal mission.
Before the studio I spent nineteen years in data. I led a twenty-person team on a national demographic statistical program for the Aga Khan Development Network in Mumbai. In Tampa I built reporting for healthcare and telecom, then spent three and a half years inside Citi's enterprise data governance, helping enforce thirteen governance policies and eight data standards across business lines.
That work taught me the thing I now apply to AI: the model can reason all it wants, but software has to control what it is allowed to do, a human has to own one decision, and the metric has to be right before you trust the reading.
I write about it on LinkedIn, teach it on SalStack, and let an autonomous agent prove it every Monday on an open-source news board with rollback rules it cannot touch.
How I work
Four things you can count on.
- 01
Direct
You work with the person doing the work. There is no account manager between you and the strategy, the words, the visuals or the build.
- 02
Evidence first
Every claim carries its source: an IRS citation on a transaction, a traceable dependency behind a mission impact, a measured metric before a fix.
- 03
Bounded autonomy
Agents get a written backlog, permission boundaries, rollback thresholds and a kill switch. They earn scope by shipping without incident.
- 04
Dated
This site says when things were true. The Now page carries a date. Notes carry dates. Products say what is live and what is in development.
Operating principles
“The model reasons. Software controls what it is allowed to do.”
- 01
Ship the agent that knows when not to.
The first time my autonomous dev agent ran alone, it could not reach the test suite. It refused to write code, logged “blocked,” and went digging instead. Best decision it has made.
- 02
Fix the ruler before the reading.
A health check said five of forty-six feeds were alive. Most were fine. The metric was broken, not the feeds. A week after fixing the metric, the same check read forty-three.
- 03
Shrink what one stolen credential reaches.
You cannot pace the frontier for anyone else. You can decide what a leaked agent token can read, when it expires, and whether thirty-four hours of abnormal volume would page a human.
- 04
One human decision, placed on purpose.
Approve, revise or reject. Everything before that gate is measured; everything after it is logged. That is the difference between an agent and a confident draft generator.
- 05
Small businesses should not receive smaller ideas.
LLCreativity is kept small on purpose. When you hire the studio you work with the person shaping the strategy, writing the words, directing the visuals and building the experience. No handoff.
Get in touch
Let's build something that holds up.
Federal cloud and security practitioners, CPA firms, founders shipping agents: I'd welcome a short technical discussion or a synthetic-demo walkthrough.
Tampa Bay, Florida · —:— local