AI engineer and founder in Cairo. Co-founder of AOM Tech, founder of InTheLine. I find where AI earns its place in a business, then build it, price it and take it to market.
I start with the problem, not the model: a founder's cash flow, a café's queue, a homeowner's repair, a patient's next meal. Then I build the AI that fixes it, price it, and take it to the market that needs it.
I co-founded one company and founded another. In both I work the business side and the build: finding the customer and the price, then shipping the AI that earns it.
Co-founder
An AI solutions startup that builds practical AI products, apps and websites for businesses and everyday users.
We take a model from notebook to production, then market it and run it. A project only moves to the next stage once the last one has proven itself.
Founder
A B2B subscription for cafés that replaces the pager with the customer's own phone, and turns the wait into return visits.
It sells monthly per branch, with nothing to install and no hardware to buy, and a café goes live in a day. I run it end to end: the product, the demo, the quote, onboarding, and training the cashiers.
Select a product to see who it's for and how it earns.
Each one starts from a business question: who has the problem, who pays to solve it, and how the product reaches them.
Beta
A financial council for Egyptian founders and SMEs. Up to eight AI advisors debate a money decision before you make it. They work from your own documents, Egyptian law and calculations done by tested code.
Each advisor answers alone first, so nobody copies the loudest voice. Then they challenge each other, the Auditor deletes any claim without a source, and the Chair gives one verdict and names the dissent. Answers come in English or Egyptian Arabic, and the Teacher explains them at your level.
Business modelB2B for founders and SMEs. Monthly plans in Egyptian pounds, from a free tier to the full eight-advisor council, with a locked price for founding members.
Built at AOM Tech: product, pricing and engineering
The Chair's verdict
Sign the contract, but ask for 30% upfront first.
5 of 6 advisors agree. Dissent: Risk officer.
Live
A free budgeting app with a goose in glasses that guards your money and roasts you when you overspend. Ten golden eggs stand for the month's budget and crack as you spend. It installs from the browser, works offline, runs fully in English or Egyptian Arabic with a right-to-left layout, and keeps every number on your phone.
Business modelB2C. Free, with no ads, accounts or servers. AOM Tech's first public app.
Designer and developer
Open Saving GoosePilots with cafés
A replacement for the café pager. The cashier taps in an order, the customer scans a QR, and their own phone rings when the drink is ready. While they wait there's a bean-catching game, loyalty beans and coupons. The owner gets live numbers: orders, average wait, busiest hours.
Business modelB2B. A monthly subscription per branch replaces pagers bought up front.
Founder
In development
A home-services marketplace for Egypt. A homeowner posts a job with photos, nearby plumbers, electricians and carpenters send competing quotes, and the homeowner picks one. Identity checks, an arrival passcode and an Arabic-first app for craftsmen are built to make strangers trustworthy.
Business modelA two-sided marketplace with open bidding instead of fixed prices.
Founding team, AOM Tech
Early build
An AI companion for people carrying something big mostly on their own. It remembers you without setup, notices when something significant happens, and checks in the next day instead of waiting to be asked. You can see, edit and delete everything it remembers.
Business modelB2C consumer app. Pricing is still being tested with early users.
Founder
My graduation project: an AI assistant for people with diabetes. Photograph a meal and it works out what's on the plate, the carbs, a dose, and the next 24 hours of glucose. The paper behind it is accepted in Informatica (Web of Science, IF 3.429).
Example output across all 101 Food-101 classes
Example plate. Error bars are the model's mean absolute error on Nutrition5K.
The base formula from the published engine (ICR 15, ISF 40, target 110 mg/dL, capped at 20 units). A demo for learning, not medical advice.
Illustrative curve. The shaded band is the 70–180 mg/dL target range.
A two-stage transfer-learning run on an EfficientNetV2-S backbone reads a photo, or a typed dish name, and ranks all 101 Food-101 classes.
86.63% top-1 and 97.43% top-5 accuracy
A regression head trained on Google's Nutrition5K predicts carbohydrates, energy, fat and protein from the same image.
Mean error of 5.18 g carbs and 60 kcal
A seven-layer dosing engine follows ADA 2024, ISPAD 2019–2024, WHO and NICE guidance: insulin-to-carb ratio, sensitivity factor, HbA1c calibration, dawn-phenomenon correction, glycaemic index and BMI tiers for Type 2. Drag the sliders to run the base formula yourself.
A BiLSTM, a CNN-LSTM with attention and a Transformer forecast glucose together as a weighted ensemble. SHAP values explain every forecast, so a patient can see why the curve bends.
An R² of 0.9859
A FastAPI and SQLite app with 18+ REST endpoints, JWT and bcrypt auth, a live patient dashboard for glucose, water and wellness, 7-day meal plans, risk alerts, and a nutrition chatbot running LLaMA 3.3 70B.
Five peer-reviewed papers from my work as a research assistant at The British University in Egypt. Most are on explainable AI in healthcare: models that predict, and then show why.
, ICCA 2023, IEEE Egypt
Abd Elrahman H. El-Sofany, Hosam F. El-Sofany, Samir A. El-Seoud
, Informatica, Q1, vol. 36, issue 9
Hosam El-Sofany, Abd Elrahman H. El-Sofany, Belgacem Bouallegue, Abdelmoty M. Ahmed, Islam A.T.F. Taj-Eddin
doi.org/10.61586/YtTX, Informatica, Q1, vol. 36, issue 10
Hosam El-Sofany, Abd Elrahman H. El-Sofany, Abdelmoty M. Ahmed, Shanawaz Ahmed J., Belgacem Bouallegue
doi.org/10.61586/CyWjG, ICL2025 and the 53rd IGIP Conference, Budapest
Hosam F. El-Sofany, Samir A. El-Seoud, Abd Elrahman H. El-Sofany
, Informatica, Web of Science, accepted
Abd Elrahman H. El-Sofany, Belgacem Bouallegue, Hosam El-Sofany
I'm an AI engineer who thinks like a founder. I co-founded AOM Tech, an AI solutions startup, and founded InTheLine. The question I keep coming back to is where AI actually earns its place in a business, and who will pay for it.
The research habit carries into business: market research before features, pricing tested with real customers, and honest labels on what is and isn't built yet.
2023
2025
2026
For B2B and B2C products, from first interview to first invoice.
From training the model to shipping the product around it.