Omar Gaafar AI workflow automation & agentic systems engineer
Builds
Agent systems · Automations · Dashboards · Websites
Depth
Parallel agents · MCP · Skills · Hooks · Memory
Work auth
Canadian · TN, no lottery

Portfolio — Omar Gaafar

AI Systems Engineer

I build the agent systems, workflow automation and operations dashboards a business actually runs on.

Open to
AI Engineer · Agentic Engineer · Automation Engineer
Based
Cairo · willing to relocate on offer
Start
Immediately

01 — What I build

Built to give time and money back.

Live operations dashboard

Every number that matters on one screen, moving as it happens — revenue, cost, margin, what's filling and what isn't.

Kills the report someone rebuilds every Friday.

Operations dashboard with revenue, cost, profit and capacity tiles, a revenue time series, and a severity-ranked attention list.

Leads & follow-up automation

Every enquiry from WhatsApp, Instagram, the website and inbound calls lands in one record, gets a real answer inside a minute, is qualified and routed, then chased on a schedule until they book or say no.

No lead goes unanswered, and none goes cold unnoticed.

Leads and follow-up automation board: capture, first-minute reply, qualification and routing, scheduled follow-up and booking, each running unattended with its last result.

Sales process automation

First message to a paid, confirmed seat — enquiry captured from every channel into one record, assessment routed, seat held, payment confirmed. One flow, running continuously.

Nobody re-types a lead into a spreadsheet again.

Sales screen: captured leads, conversion and enrolments, above an automated end-to-end sales process marked running automatically.

Anomaly & loss detection

Every branch and every line measured against its own history and against its peers, ranked by severity, each finding linked to whoever can act on it.

Finds the leak while it's still small.

A severity-ranked attention list: expiring holds with money at risk, borderline evaluations held for a person, and programmes falling month on month.

AI finance agent

One-click reports that shift with what finance actually asks for, and an agent you can ask anything — it answers in a sentence, a chart or a table, whichever the question deserves.

Turns a two-day analyst request into a ten-second one.

Finance agent with one-click report cards, a natural-language question box, and margin and outstanding-balance tiles.

AI accounting agent

Invoices, settlement reconciliation, expenses and refunds — asked in plain language, answered with the working shown so an accountant can check it rather than trust it.

Removes the data entry, keeps the judgement.

Accounting agent with one-click reports for monthly export, payment reconciliation, outstanding and ageing, and expense breakdown.

Customer support agent

Answers drawn from the business's own material, running day and night on the website and the channels customers already use — handing over cleanly when a person is genuinely needed.

Catches the enquiry that arrives at 2am.

A live website AI agent, online, with suggested opening questions and a free-text box.

HR & hiring automation

Contracts issued and sent on signature, shift rotas built from the contract, training dispatched, attendance chased — running unattended, with a manual override on every one.

Time-to-hire measured in days, not weeks.

HR automation board with six automations, four running unattended, each showing its last run and result.

Approval & request workflows

Typed intake per department, routed by rule, ordered by severity, with dual confirmation wherever an item can't be closed on one person's word — and every stage owned by exactly one role.

Nothing sits in a message thread for eleven days.

An approval queue showing all clear, above the six-stage request flow and the average time spent between each stage.

Plus websites and product interfaces — like the one you're looking at now. Every screen above is a system I built and ran; client branding has been removed.

02 — Agent systems

01 / 08

Agents that do the work, not just the talking.

I run a twenty-agent system daily and built every part of it. These are the capabilities that separate an agent that can act on a business from a chat window with a good prompt.

  1. 01

    Parallel agents

    Independent work runs at the same time instead of queueing — several agents on different parts of a problem at once, each in its own context, reporting back to be reconciled.

    Why it matters — the difference between an answer in two minutes and one in twenty.

  2. 02

    Delegation by task

    Any agent can hand work to any other, so the chain forms around the problem rather than being fixed in advance — with rules that stop a delegation loop before it starts.

    Why it matters — a rigid org chart makes the system slower than the person it replaced.

  3. 03

    Skills — and writing them

    Packaged procedures an agent loads only when the task calls for them, so expertise is available without sitting in memory permanently. I author them, and I built the skill that writes new ones.

    Why it matters — a process stops depending on whoever remembers how it goes.

  4. 04

    MCP servers

    Model Context Protocol servers give an agent typed, permitted access to what a business actually runs on: the database, the document store, version control, the file system.

    Why it matters — an agent that can't touch your systems can only ever give advice.

  5. 05

    Cross-session memory

    Decisions and prior work persist between sessions in a searchable local store, so the system starts knowing what was settled last week instead of starting from nothing.

    Why it matters — re-explaining the business every morning is why most agent pilots quietly die.

  6. 06

    Hooks — hard guardrails

    Ordinary code running before and after every action, outside the model's control: blocks destructive commands, keeps credentials and personal data out of anything that leaves the machine.

    Why it matters — safety you can only ask for politely isn't safety.

  7. 07

    Model routing

    Mechanical work runs on the cheapest capable model; only ambiguous or irreversible decisions escalate. The routing table is a real design document, not a default.

    Why it matters — this one decision is what makes a system this size affordable to run.

  8. 08

    Scoped tools, audited output

    Each role gets only the tools its job needs, and every result is treated as a claim to be checked — a number without the query that produced it doesn't get used.

    Why it matters — a confident wrong answer costs more than no answer.

03 — Value

What a business gets out of it.

04 — Stack

What I work in.

Agent runtime

  • MCP servers
  • Skills, and authoring them
  • Hooks & guardrails
  • Delegation graphs
  • Model routing
  • Cross-session memory
  • Parallel execution
  • Scoped tool grants
  • Structured outputs
  • Tool calling

Automation

  • State machines
  • Approval routing
  • Role-based access
  • SLA & breach detection
  • Notification pipelines
  • Event-driven jobs
  • Scheduled jobs
  • Webhooks
  • System integration

Retrieval

  • Vector + keyword + graph
  • ChromaDB
  • Neo4j
  • Knowledge graphs
  • Chunking strategy
  • Query expansion
  • Re-ranking
  • Context budgeting

Languages

  • Python — asyncio, Pydantic
  • TypeScript
  • SQL
  • Java
  • Bash

Interfaces

  • Next.js
  • React
  • React Native
  • Tailwind
  • FastAPI
  • REST
  • WebSockets

Reliability

  • Benchmark harnesses
  • Per-stage instrumentation
  • Baseline comparison
  • Threat mapping
  • Prompt-injection defence
  • Least-privilege scoping
  • PII handling
  • Git

Data

  • Pandas
  • NumPy
  • ETL pipelines
  • Data modelling
  • Power BI — DAX, Power Query
  • Statistical analysis
  • Scikit-learn
  • TensorFlow
  • Neural networks
  • NLP
  • Classification & clustering
  • Model evaluation

05 — Track

Where I've done it.

  1. 2026 —Founder & AI Engineer, Moga — AI automation studioCairo
  2. 2026Bachelor thesis, AI Engineering — with AiCentive GmbH. MCP Enterprise Productivity System with HybridRAG and Knowledge GraphsDHBW Stuttgart · A+
  3. 2025–26AI Engineering & Software Development InternInternational House Cairo
  4. 2025Software Development Intern, Frontend React Native teamArab African Int. Bank
  5. 2025IT Intern — network, cybersecurity, webElSewedy Electric
  6. 2025Business Intelligence InternBingo Global Group
  7. 2022–26B.Sc Business InformaticsGerman University in Cairo

06 — Contact

Hiring for this?
Let's talk.

omarggaafar@gmail.com
Phone
+20 10 1032 9954
Open to
AI Engineer · Agentic Engineer · AI Automation Engineer · Forward-deployed
Citizenship
Canadian
US work authorisation
TN under USMCA — Computer Systems Analyst. No annual cap, no lottery, no PERM
Employer cost
None. Processed at a port of entry, or premium filing in 15 days, on offer
Location
Cairo · willing to relocate on offer
LinkedIn