Operations · Programs · Automation
I build the operating layer behind subscription and fintech businesses, and the software that runs it.
Seven years of UAE operations across government service, healthcare, e-commerce and subscription technology, turning fast-moving products into systems that hold: customer journeys, SOPs, risk frameworks, and the reporting leadership actually decides from. Then I got tired of waiting for tooling and started shipping it myself.
Selected work
A 27% default rate, and every approval argued case by case
A weighted scoring model, agreed thresholds and a quarterly refresh, built so finance, risk and commercial approved against the same numbers instead of arguing case by case.
Corporate onboarding took two weeks and nobody owned it
Documents, credit checks, agreements, e-signature and direct debit stitched into one sequence with a named owner at every step. The speed came from deleting steps rather than automating them.
A live product running on instinct, with no process written down
The full device-subscription journey designed from a blank page across delivery, device capture, warranty claims and reverse logistics, with retail partners in the Apple ecosystem.
A cross-border launch that did not need launch-day firefighting
Readiness run a week early across logistics, customer service and returns, then the service model rebuilt around what the complaints were actually saying.
Work with me
Two ways in: a senior operations or programs role, or a scoped project. Projects are remote, in Arabic or English, for teams across the GCC and East Africa. Each service below is something I have already run in production, with the case study linked as proof.
Operations design for subscription, fintech and e-commerce
Onboarding flows, credit-risk decisioning, SOPs and customer journeys, each step with a named owner and one number that operations and leadership both accept.
AI automation with a human approval step
Workflows on the Claude and OpenAI APIs that draft, check and route for approval instead of acting alone, with an alert when a scheduled run fails silently.
Proof: 170 sources, nothing sent unseen · amount-verified payments
Arabic document and data pipelines
Arabic OCR, extraction and verification gates that turn posts, scans and exports into a searchable registry, at a running cost low enough to keep it going.
Proof: 679,162 records, AI spend ÷10 · 7 export formats, one corpus
Write to me with what is breaking and the number you want to move. Firasfmohammed@outlook.com
Built on my own time
I have no formal coding background. What I have is an operator's instinct for where a process is bleeding, and enough tolerance for friction to keep going until the thing runs. These are real systems, not demos: a computer-vision platform, a product with live billing, a company I registered and run remotely, a recovery database serving families I have never met, the job-search engine I rebuilt around my own pipeline, and an ETL pipeline over seventeen years of archives.
Idle site cameras turned into a searchable record in three weeks
Incumbents charge from USD 10,000 a project; most sites already own cameras nobody reviews. ConstructVision makes that footage searchable in plain language, with the access controls, tenant isolation and release gate a paying customer audits before signing. Staging, pre-customer, stated plainly.
JobVerdict: a paid product where every payment grant is amount-verified
A SaaS is only sellable once money moves through it correctly. Multi-tenant, metered per user, plan quotas, live UAE payments with idempotent, amount-verified grants, and an anti-fabrication guardrail enforced in code and covered by its own tests. Designed, directed and shipped end to end.
679,162 free public records from the chaos of Arabic social media
Families were searching post by post for one chassis number. The pipeline turned 106,595 posts and 133,161 photos into a searchable registry, cut its AI spend tenfold so it can stay free, and caught a 130,000-record silent loss that the completion metrics had passed as clean.
Founding a company in Sudan and running it from Dubai
An events and catering business registered, banked and trading, with a partner team across three countries that I manage remotely. I built the invoicing and verification system it bills on, and the reporting the partners actually read.
An open-source job-search engine made safe to run unattended
I treated my own job search as an operations problem: 170 verified portals, a triage dashboard, status read from the mail store instead of fragile scraping, and an outreach queue that drafts, holds a veto window and routes approvals to my phone. The control model I would put around any automation that touches customers.
Seventeen years of archives, seven formats, one queryable corpus
Every platform will export your data; none will hand it over in a shape you can use, and one corrupts Arabic on the way out without reporting an error. Seven incompatible exports normalised into a single store, with a runbook written for someone who isn't me.
What these builds show about how I run things
- Information management. Chaotic, multilingual public data turned into a structured, searchable registry: 106,595 posts and 133,161 media items became 679,162 records anyone can query.
- Data integrity and reporting accountability. Four verification gates on every processing wave. One of them caught a silent loss of about 130,000 records that the completion metrics had reported as a clean pass. Verify the records, not the metric.
- Access governance and data privacy. Role-based access, tenant isolation, MFA and session revocation designed in from the first release, so one organisation's data never leaks into another's.
- Cost stewardship. Model spend cut by an order of magnitude through benchmarking, deduplication, model tiering and batch processing. Measure before paying to scale.
- Release and quality gates. One command runs tests, typecheck, lint, build and migration checks before anything ships, mirrored in CI. Payment grants are idempotent and amount-verified, and the anti-fabrication rule is enforced in code and covered by its own tests.
- Human-in-the-loop controls. Automation that drafts rather than sends, holds a veto window, routes approvals to a phone and alerts on every failed scheduled run.
- Risk-based roadmapping. An independent teardown of the platform's own limits became the roadmap, and an operational audit that found 23 defects was sent back when the recheck was not ready to sign off.
Notes
Short write-ups from the systems above: what broke, what it cost, and the number that changed.
How I cut the AI cost of an Arabic OCR pipeline about tenfold
The cost layers of a pipeline that turned 106,595 Arabic posts and 133,161 photos into records: free checks first, OCR once, model tiers, batch pricing.
B2B customer onboarding in the UAE: from 2 weeks to 5 business days
The UAE B2B onboarding sequence I ran, from trade licence to direct debit, and how one owner per step cut it from 2 weeks to 5 business days.
Your construction site cameras already record everything. Can you search it?
Most building sites already record every day on fixed cameras. What plain-language search over the footage takes, and the access checks a buyer should run.
How to check if your looted car in Sudan has been found
Arabati Wein (mycarsd.online) is a free registry of cars looted in the Sudan war. How to search it by chassis or plate number, and what to do with a match.
How I work
- Ask what can be cut before asking what can be automated. Most cycle time is not work, it is waiting at a handoff that belongs to nobody. Automating that just makes the wrong process faster.
- One number both sides accept. A metric that operations and leadership have agreed on does more for accountability than a governance document nobody reads.
- Standardise the 80% that repeats, protect the 20% that does not. Over-standardising quietly kills the exceptions that matter most to the customer.
- A process without a named owner and a sign-off is a habit. It survives exactly as long as the person who remembers it.
- Measure before you pay to scale it. My most expensive architecture decisions were the ones I made before benchmarking. Every one of them got cheaper once I did.
Where I have done it
- Zension Technologies (Zaam)
Manager, Programs, Operations & Portfolio Risk · device subscriptions, UAE · Aramco and Sumitomo backed · 2024 to 2026 - ZEN VIBE
Founder · events and catering, Sudan, operated remotely · 2026 to present - Revibe.me
Business Operations Senior Specialist · cross-border e-commerce, Dubai and Kuwait · 2023 to 2024 - German Neuroscience Centre
Business Operations Team Leader · multi-branch private healthcare, Dubai · 2021 to 2023 - Dubai Electricity & Water Authority
Customer Service Team Leader · public service delivery under the Dubai Government Service Model · 2019 to 2021
Skills & toolkit
Contact
Open to Head of Operations, Operations Manager, programme and vendor-operations roles across the UAE and GCC, to programme, operations and information-management roles with UN agencies and international organisations in Nairobi, and to scoped remote projects in operations design, AI automation and Arabic data.
I read everything that arrives, and I answer.