I build AI agents people actually use, from the first idea to daily work, with guardrails and real adoption. Startup, project, discovery or research: let's build it.
Local government, credit data, then retail at scale. Same job each time: find where the process breaks and fix it.
Double Star Ranger, 209 badges, 24 trails
Agentforce Service, Prompt Builder Templates
Innovator 2026
SAFe 4 Certified Scrum Master
B.S. Computer Information Systems, UMGC (in progress). A.S. Computer Science, College of Southern Maryland
Apex, Lightning Web Components, Flow, Visualforce, integrations, Copado CI/CD
Model Context Protocol, LLM evaluation, model routing, Claude Code, GitHub Copilot
JavaScript and TypeScript, React, Node.js, Python, pandas, Streamlit, FastAPI, SQL and SOQL
GitHub Actions CI/CD, architecture decision records, test-driven delivery, , and governance
Work builds are described in general terms: no internal names, data or code.
An SDK and CLI that turn architecture-review standards into automated code review. It grounds GitHub Copilot and Claude Code in the review rule catalog, checks only the rules that apply to each component, and pulls in story data and runtime logs.
Reviews in 1 to 2 minutes. Rolled out to 140+ engineers. First release in 8 days.
A Salesforce-native engine that replaced a manual spreadsheet process. It scores service providers daily and traces every figure to its source records.
Cleared , and in a single week.
A Salesforce self-service app that replaced scattered emails and calls with one tracked, routed workflow for workforce requests.
60,000+ requests a year. Sole developer since 2022.
Service-provider onboarding with e-signature and a nightly license validation job that flags expired documents for the team. It informs people and never acts on providers automatically.
Nearly 5,000 labor hours saved per year, validated by the business.
A Model Context Protocol server that gives AI agents one audited way into business systems, with central credentials and a built-in personal-data block.
Matched each AI agent to the right model tier, moving agents to a lower-cost tier wherever the job allowed.
Real-time view of where every vendor stands, a worklist of the oldest open items and row-level access.
A nightly job flags each case and logs a follow-up for the team, replacing a manual review queue. It informs people and never acts on vendors.
Audits users, flags expiring access keys and reconciles accounts against the CRM. Also designed a pilot extending digital signing to more of onboarding.
Merged and grouped license records and surfaced likely duplicates for review.
The first baseline for a support channel: years of cases, with contact reasons classified by privacy-preserving rules.
Org switching with production safeguards, a query and export workbench, and a bulk importer that maps fields live. CI-gated release.
Rubric scoring, duplicate checks and transparent effort estimates. Every project tracks expected value against measured value.
A privacy-first browser app that turns a learning-system export into past-due views by leader and drafts follow-up emails. No data stored or sent.
An unstructured shared inbox turned into categorized, routed cases with automatic status replies and reporting by category and owner.
Tracks design reviews from assignment to approval, generates barcodes and produces printable PDF review packets, through four environments.
A certificate generator, an award-form print tool, a looping office announcement display, a searchable site of every team presentation, and interactive team presentations.
How fast I go from idea to something people use, with where every number comes from.
Service-provider onboarding automation with e-signature and nightly license validation.
Through a Salesforce self-service app I built and run as sole developer.
Automated architecture-review checks in 1 to 2 minutes instead of manual prep.
Then more than 70 releases in 8 months through a gated dev-to-prod pipeline.
07Approved and released. Review prep took 1 to 2 minutes.
The same loop whether it is a kids app, a client demo or a production change at work.
Before anyone gets asked a question, my agents read the project notes, the data and the history. Most answers are already written down somewhere.
Fast, cheap models for bulk checks. The strongest model at high effort for design and architecture. Standardizing this moved agents to a lower-cost tier wherever the job allowed.
Design and interaction need judgment. Worth the extra time to get the first impression right.
Each agent gets one job, a clear brief and guardrails, and logs what it decided, so any result can be audited. Nothing goes out without a human approving it.
Click every button, check that the numbers make sense, screenshot the result. Then the same release governance as any production change: CI gates and architecture, implementation and change review.
A tool only pays off when people use it. I gave a talk on structuring AI coding workflows, wrote my team's prompt-engineering and code-review standards, and coach leaders and teams on adopting Claude.
What context to hand over, what to leave out, and how to say what done looks like.
Where AI output usually goes wrong, and the quick checks that catch it before anyone else does.
Which steps still need a person, and how to tell before you spend a week building.
Startups, projects, discovery, research or teaching. If it involves AI agents, automation or a process worth fixing, I want to hear about it.
I get an idea in front of people fast, then keep shipping. Baby Play went from idea to v0.1.0 in one day; my AI agent SDK shipped its first release in 8 days, then more than 70 releases in 8 months.
Agents with one job, approved sources, guardrails and a person in the loop, taken through review and into daily use. My automation work saves nearly 5,000 labor hours a year.
I start with what's already written down, ask the discovery questions, and map which steps an agent should take and which stay with a person, before anyone spends a week building.
Model routing and LLM evaluation: matching each job to the right model tier, and testing with deliberate traps before anyone relies on the result.
Talks, prompt-engineering and code-review standards, and coaching for leaders and teams adopting Claude.
Describe a business problem in a sentence or two. The AI version of me outlines how I would take it from discovery to an agent in daily use, using my real method. It shows how I think; it is not a quote or a promise.
What I build outside work. Some run privately or were delivered to one person; Baby Play and Aghani are finalizing for launch.
LLM agents with defined roles, tools and prompts that triage email, prepare meeting briefs and watch topics on a schedule. Connected through MCP to email, calendar and a Next.js dashboard, with human approval before any outbound action.
A network admin needed to plan VLANs and subnets without a spreadsheet. An interactive planner with an editable topology.
A SwiftUI app for toddlers: pop balloons to learn letters, numbers, Arabic letters and animals. Idea to v0.1.0 in one day.
Arabic kids songs in Egyptian Arabic. The first three songs were built on day one.
Hussein AI greets you on the page and follows you as you scroll. It is also a working sample of how I design agents: a narrow job, approved sources and an honest handoff.
It asks who you are first, so a recruiter, an investor and a business owner each get a different path.
See howOnly my case studies, build log and experience. If it isn't written down, it says it doesn't know.
See howIt points you to me on LinkedIn for anything it shouldn't answer. No pretending a message was sent.
See how| Compare | About page | Hussein AI | |
|---|---|---|---|
| Tailored to the visitor | Same page for everyone | Same profile for everyone | A path per visitor |
| Shows the work | Bullet points | Bullet points and posts | Live demos and receipts |
| Answers your question | No | Only by message | Right away, any hour |
| Says when it doesn't know | Not applicable | Not applicable | Yes, then hands off to me |
| Proves the skill | Claims it | Claims it | Is the skill |
A few things that have nothing to do with work.