Usama Ahmed. AI Engineer and Forward Deployed Engineer. I work inside your team, from the first requirements meeting to a system running in production.
requirement gathering
I start with the people who own the problem, not with a model. I keep asking until the real requirement is on the table.
solution and architecture
Before any code, the whole system fits on one page: what each agent does, what it may touch, and where a person decides.
agents and development
An orchestrator plans. Small agents do one job each: retrieve, call a tool, check the result. Every hand-off is logged, so you can see why the system did what it did.
ci/cd
Nothing reaches production by hand. Every change passes the same four gates, and prompts are versioned like code.
monitoring, guardrails, governance
Going live is the start. I watch quality, cost and drift, and put guardrails where a wrong answer is expensive. Risky cases go to a person.
outcome
Not a demo. A system your team runs.
That is the whole job. From the first conversation to production, with your people in the room.
about
I started as a full-stack engineer and spent the last seven years deep in frontend architecture with Angular. Along the way I built and ran three startups and led a team of more than ten people. That taught me the part most AI projects skip: talking to the people who own the problem, and staying until the thing works.
Why AI now. Agents are finally good enough to do real work, but only inside a well-built system: requirements, architecture, evaluation, operations. That is engineering, and it is what I do.
experienceplaceholder: dates and company names
[20XX to now]
[Company], Nuremberg
Frontend architecture for [product]. Brought AI agents into the delivery workflow: [one concrete result, with a number].
[20XX to 20XX]
[Startup three]
Built the product from zero and led a team of more than ten. [What it did, who used it, how it ended.]
[20XX to 20XX]
[Startups one and two]
Two earlier companies. [One line each: the product, the customers, the lesson.]
[20XX to 20XX]
[Company]
[Where it started: the stack, the kind of systems, one result.]
work
A VS Code extension that turns an AI agent's plan into steps you can see, run and stop.
Add rate limiting to the public API
A retrieval assistant that drafts ticket replies from four internal sources, with a source on every claim and a human in the loop for refunds.
A regression suite that scores every prompt and model change before it ships, and blocks the release when quality drops.
stackplaceholder: assumed stack, confirm every item
What people touch.
The part that thinks, and the part that checks it.
APIs and data.
How it reaches production and stays there.
Where it runs.
writingplaceholder: sample articles, layout only
A plan you cannot inspect is a prompt you cannot debug. What changes when execution is step by step.
How a regression suite for prompts became the release gate on every project, and what it caught.
Whom to talk to, what to write down, and what not to build yet.
contact
Tell me what is slow, manual or stuck. You get a straight answer on whether AI helps, and what it would take.