One person.Whole stack.
I'm Zain. Database, agents, API, interface. I would rather hand you something running than a diagram of it.
Why I work this way
I started working at 15, writing esports journalism, and by 17 I was doing investigative reporting at a publication that had just been acquired. I never stopped after that. I did my undergrad in management information systems and a master's in business analytics at the University of Georgia while working the entire way through, journalism for the first two years and AI for the last two.
At 19 I joined GreenSky's AI governance department the week it was stood up, absorbed three roles over two years, and held the technical picture for the group. When that finished I went the independent route instead of the standard one, because I wanted to keep building the things I was actually interested in rather than the next item on somebody's roadmap.
The through-line in all of it is a bias toward the practical. You can plan for a lot, and then you hit a ceiling. The things that actually sink a solution only surface once somebody implements it: the schema / the retrieval / the one API call nobody read the docs for. So I implement early, on purpose, while those are still cheap to find.
That is also what two years in governance taught me. Most organizations have a real gap between how they talk about AI and how AI actually works, and almost nobody standing in that gap can explain both sides. I learned every layer because of it. If you can only touch one layer, you spend the project negotiating with the layer that is actually broken.
How I work
Before anything gets built. Most projects are late because the team solved a stated problem instead of the actual one. I would rather spend the first week making you defend the brief than spend the last month rewriting.
I default to owning the pieces that hold your data instead of renting them. Vendors change pricing and terms. A system you own does not get repriced, and it does not leak your context into somebody's training set.
Every solution gets trimmed to what the problem actually needs. Extra architecture is not free, you pay for it every time someone touches the code later. The cheapest thing I can hand you is something you can still change.
You get the code, the deploy path, and the reasoning behind both. Nothing I build should require me to keep working for you. If the only person who can change it is the person who wrote it, it was built wrong.
Things you can go check
pgvector, running on hardware I own rather than in a cloud, reached over a private network, with an MCP server in front so Claude can query it directly. It holds real contact data, so it does not live where the website lives.Work I can talk about
I build constantly, and most of it is my own: experiments, prototypes, tools I needed for a week and then kept. That is how I stay current when the tooling changes every month. The client work is a slice of that, and most of the slice is private. These two I can describe.
The review that only worked if the right person remembered
I worked at GreenSky, in the governance department, and this was the last thing I built there. Every change the company wanted to make, a new system, a new vendor, an AI feature switched on somewhere, had to clear a review. Whether it cleared depended on who was in the room and what they happened to recall. Which controls already covered it. Whether a policy needed rewriting. Whether the same question had been asked and answered eight months earlier. That knowledge sat in people's heads, one department at a time, and there was no way to ask it a question.
I took all of it and made it searchable against the meeting transcripts, then built the checks on top. A proposed change now comes back with:
- Controls Which ones already cover this, and whether the change reconfigures any of them.
- Policies and procedures Whether this introduces something that affects them. If it does, it goes on the next review cycle, with renewal dates tracked. Policies get redone yearly, procedures run on no fixed clock, so those were tracked separately.
- Regulation Whether it lands near existing or proposed AI rules, or FTC territory. If it does, it gets flagged before anyone builds it.
- History Whether this topic has been through review before, what was asked at the time, and which follow-ups are still open.
- Vendors Whether it is a new vendor, or an existing one turning on a new capability like AI, which has to trigger their own pipeline.
The gaps it closed: AuditBoard controls drifting out of date, and new vendors getting through without anyone catching them. It is the reason they kept me on.
Anyone can paste a contract into AI. Nothing tells you whether the answer is right.
A scope of work decides whether a project is profitable or a nightmare, and most independent consultants sign them without a lawyer. You can drop one into a chatbot today and it will rewrite it for you. What it cannot do is tell you what normal looks like, because it has no idea what consultants in your position actually charge, what payment terms are standard, or which protections yours is missing. There is nothing anchoring the answer.
This is anchored to real consulting data. The pricing feedback is a comparison against what the work actually goes for. And it does not just hand back a rewritten document. Every flag comes with why it was raised and what the scope is missing, which leaves you able to decide what to fight for. A rewrite takes that decision away from you and hides it.
The scoring runs in code rather than through the model, so the same document gets the same answer twice. The redlining was the hard part underneath all of it: a model can tell you a clause is dangerous and still not know where that sentence physically lives inside a Word file, and guessing wrong corrupts somebody's contract. Every edit is verified against the real authored text before anything gets written.
What it is like to work with me
Direct. I will tell you when I think the plan is wrong, once, with the reason, and then I will build what you decided. I do not perform certainty I do not have, so if I say something is a guess it is a guess, and if I say it is verified I checked.
You will always know where a thing actually stands. Nothing gets called done because a deadline says so. If something is half built I will tell you which half, and what it would take to finish the rest.
p.s. I am not interested in AI theater. If a tool does not survive my own week of using it, I do not ship it to you.
Got something that needs building properly?
Tell me the problem, not the spec. If it is a fit I will say so quickly, and if it is not I will tell you that too and probably point you somewhere better.