01 Live Ops
Revenue, profit and every problem in the business — live, on one screen
- 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
Twenty automations. Two agents. One screen to run the business.
- 01Live Ops
- 02Sales
- 03Customer Service
- 04Finance
- 05People & Admin
- 06Marketing
- 07Anomaly Detection
- 08One-Click Reports
02 Sales
Quotes written, sent and chased until they close
03 Customer Service
One agent answers every channel in seconds, and hands off when a person should step in
04 Finance
An agent that answers any question about your money, straight from your own books
05 People & Admin
Offer letters, payslips and contracts, written for you
06 Marketing
Posts drafted, comments answered, reviews asked for and replied to
07 Anomaly Detection
Every number watched — the problem flagged, with the reason, before it costs you
08 One-Click Reports
The P&L, cash flow and VAT return, built in one click
The real platform, running on sample data — not a client deployment.
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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
Hours back
Automation removes the repeated work, not the judgement
Intake, routing, chasing, scheduling and reporting are rules dressed up as jobs. Once they run themselves, the people who were doing them go back to the parts that actually need a person.
From — sales, HR and approval automation
-
More output, same headcount
Agents work in parallel, and don't stop at five
Several agents working different parts of a problem at once, on a chain chosen by the task, means the throughput of a small team without the coordination cost of a large one.
From — the agent layer
-
Money kept
Losses get found while they're still small
Every branch compared against its own history and its peers, ranked by severity. The outlier surfaces itself instead of waiting to be noticed in a quarterly review, which is usually one quarter too late.
From — anomaly and loss detection
-
Faster, more informed decisions
The answer arrives before the meeting, and it is the right one
When the numbers are already on screen and any question can be asked in plain language, the delay between wanting to know something and knowing it stops being measured in days.
From — live operations and the finance assistant
-
Nothing dropped
Every item has an owner, a clock and a next step
Severity ordering, breach detection and dual confirmation mean the question "did that ever actually get done?" has an answer that doesn't require asking anyone.
From — approval and request workflows
-
Always open
The business answers at 2am
Enquiries captured and answered from the business's own material across every channel, day and night, with a clean handover the moment a person is genuinely needed.
From — the support chatbot and sales automation
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
- Document generation
- Anomaly detection
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
- Radix
- FastAPI
- REST
- WebSockets
Data
- Pandas
- NumPy
- Power Query (ETL)
- Data modelling
- Power BI — DAX
- Statistical analysis
- Scikit-learn
- TensorFlow
- Neural networks
- NLP
- Classification & clustering
- Model evaluation
05 — Track
Where I've done it.
Experience
- 2026 —Founder & AI Engineer, Moga — AI automation studioCairo
- 2025–26AI Engineering & Software Development InternInternational House Cairo
- 2025Software Development Intern, Frontend React Native teamArab African Int. Bank
- 2025IT Intern — network, cybersecurity, webElSewedy Electric
- 2025Business Intelligence InternBingo Global Group
Education
- 2026Bachelor thesis — AI EngineeringMCP Enterprise Productivity System with HybridRAG and Knowledge Graphs, with AiCentive GmbHDHBW Stuttgart, Germany · A+
- 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