S-01
Ask-your-documents assistant
An assistant that answers questions from your company's files, so your team and customers don't have to search.
RAG, LangChain, ChromaDB, FAISS
Ask about thisI build AI that reads documents and understands images.
I'm Ahmed Ehab, a forward deployed AI engineer in Egypt. I work alongside your team to turn a business problem into a working system: I design it, build it, test it, and keep it running in production.
Recent work: assistants that answer questions from a company's documents, models that read construction plan drawings, and software that reads medical scans.
What I can build for your business
Six kinds of project I have built or taught. Pick the closest one, or describe your own.
S-02
Computer vision for technical drawings
Models that find and read elements in construction plan drawings and return the results to your systems in real time. Built for construction-tech clients.
YOLOv8/v10, FastAPI, Docker
Ask about thisS-03
AI workflow automation
Connect your tools so repetitive work, like sorting requests or filling in records, runs on its own and asks a person only when needed.
n8n, tool calling, Pydantic
Ask about thisS-04
Case and document management
A web tool for your team to track cases and files in one place.
TypeScript, FastAPI
Ask about thisS-05
Medical and scientific image analysis
Models that read slides and scans, such as pathology images and brain scans, to support the experts who make the decision.
PyTorch, OpenCV, scikit-learn
Ask about thisS-06
AI and Python training for your team
Hands-on courses, from your first lines of Python to machine learning. I have taught 500+ students and MSc classes.
Python, machine learning, FlutterFlow
Ask about thisHow a project runs
Five steps, from the first call to a system in production.
- 1
A discovery call
We talk about the problem, who will use the result, what data you have, and how you will know it worked.
- 2
Scope and plan
I write down what will be built, in short sprints with measurable delivery targets, and list the risks, including data privacy and who may see what.
- 3
Build in small steps
I design and build in short sprints you can review, so you can change direction early.
- 4
Test against the targets
I measure the system against the targets we agreed, not against how good the demo looks.
- 5
Deploy and keep it running
I deploy it to production, hand it over, and keep it running for real users.
Privacy and data governance
I design systems to align with the rules your business works under. Final compliance sign-off stays with your legal and security team.
Saudi PDPL and the EU GDPR
I build with the principles of the Saudi Personal Data Protection Law (PDPL) and the EU GDPR in mind: collect only what is needed, keep it for a defined time, and respect access and deletion requests.
Your data stays where you want it
I can run open-weight models locally with Ollama, so sensitive documents never have to leave your own servers or your country.
Access control and an audit trail
Who may see which documents, and a log of who asked what.
Handling personal data
Detecting and masking personal data, such as names and ID numbers, before it reaches a model or a log.
A person makes the decision
For legal, medical, and other high-stakes work, the system supports a person's decision and shows its sources instead of deciding alone.
Tested before it is trusted
I measure accuracy and failure cases against an agreed target, and use structured outputs so answers can be checked.
Other Gulf data-protection laws and common security practices can be added to the plan on request. I am an engineer, not your legal adviser.
What I do
I build AI that works inside real businesses: assistants that answer questions from a company's documents, and the web services that put them in front of users. I work as a forward deployed engineer, embedded with the client's team, and I own the whole job: discovery, technical proposals, architecture, deployment, and measuring whether it works.
Before language models I worked on computer vision and medical data: analyzing huge pathology slides, detecting early signs of Alzheimer's from brain scans, and building object-detection models for construction plan drawings. That is why I care about what the data looks like, not just what the model says about it.
I work where "mostly right" is not good enough, so I test against agreed targets and keep a person in charge of the decision. I have also mentored engineers and introduced an AI-assisted engineering workflow to my team.
Experience
-
June 2025 – present
AI Engineer & Product Owner LOCTech, Saudi Arabia, remote
- Own the AI product roadmap and lead engagements from discovery to production: run requirement sessions with client stakeholders, write technical proposals, PRDs, BRDs, and SRS documents, and turn them into scoped sprints and acceptance criteria.
- Built and deployed production computer-vision pipelines (YOLOv8/v10, FastAPI) that serve real-time inference for construction-tech clients.
- Stood up a self-hosted automation and retrieval stack (n8n, Qdrant, PostgreSQL with Docker Compose) and GitLab CI/CD with a self-hosted runner for Laravel and Next.js production deployments, with health checks and least-privilege deploy permissions.
- Delivered an internal HR and training platform for managing employees and courses, built with AI-assisted development and reusable service patterns.
- Introduced an AI-assisted engineering workflow (Claude Code, MCP) across the tech team and designed an internal AI document generator for proposals, contracts, and PRDs with DOCX and PDF output.
-
July – September 2025
Machine Learning Researcher Intern Nile University, Egypt, hybrid
- Built an end-to-end machine-learning pipeline for diagnosing chronic kidney failure: data preparation, model selection, and evaluation.
-
March 2024 – September 2025
Technical Instructor iSchool, Rocket Academy, Metastatic ME, Egypt
- Taught 500+ students Python, computer-science basics, AI, and game development at iSchool, as part of the DEMI initiative led by Egypt's Ministry of Communications and Information Technology.
- Taught MSc students data analysis, machine learning, and bioinformatics (Metastatic ME), delivered FlutterFlow mobile-app training, and helped design the curriculum (Rocket Academy).
-
2021 – 2025
BSc Computational Biology & Genomics Zewail City of Science and Technology
- Specialized in computer vision. GPA 3.26 out of 4.0.
Selected work
Open a row for the details.
-
Models that detect and read elements in construction plan drawings, served as a real-time service for construction-tech clients.
- A production pipeline that serves real-time inference to the client's systems.
- I owned it from the client's requirements through architecture to evaluation.
- Client and project details are confidential.
YOLOv8/v10, FastAPI, Docker, CUDA
-
A chatbot that helps people write and improve their CVs, built end to end for a startup client.
- Built the TypeScript web front end and the FastAPI back end that calls large-language-model APIs.
- Delivered end to end, from the client's idea to a working product.
TypeScript, FastAPI, LLM APIs
-
Turns one or two Mac minis into a self-hosted AI service for a small team: a chat app, an OpenAI-compatible API, and per-team token budgets.
- A ChatGPT-style chat app (Open WebUI) and an OpenAI-compatible API (LiteLLM), running on your own hardware.
- Per-team keys and token budgets, with an optional cloud fallback through OpenRouter.
- Models run with Ollama on each Mac. An optional Cloudflare Tunnel gives a public address with no open ports.
- Set up from a few small config files with one command-line tool, and every command is safe to re-run.
- View minihub on GitHub
Ollama, LiteLLM, Open WebUI, OpenRouter, Cloudflare Tunnel, Shell
-
A decision-support classifier that reads gigapixel pathology slides, trained without labeling every region by hand.
- Prepared the slides: cut them into patches, separated tissue from background, normalized color, and augmented the data.
- Extracted features with ResNet18, ResNet50, VGG16, and VGG19, and combined them with multiple-instance learning.
- Trained and tuned SVM, logistic regression, decision tree, random forest, KNN, and naïve Bayes classifiers.
Python, PyTorch, scikit-learn, NumPy, CUDA, TIA-Toolbox
-
Machine learning on brain-scan measurements to flag early Alzheimer's.
- Extracted shape and texture measurements from scans with PyRadiomics.
- Normalized the features, removed highly correlated ones, and encoded the labels.
- Compared SVM, logistic regression, decision tree, random forest, and KNN on accuracy, precision, and F1.
Python, PyRadiomics, SimpleITK, scikit-learn, pandas, seaborn
Skills
AI assistants and language models
Search and data
Using and hosting models
Software engineering
Deployment and infrastructure
Machine learning and computer vision
Coding with AI agents
Languages
Beyond the code
Teaching. I have taught more than 500 students Python, computer science, AI, and game development through Egypt's DEMI program, trained MSc students in machine learning and bioinformatics, and built mobile-app courses.
Speaking. At Dakahlia STEM School I coached a TED-Ed team of fifteen in public speaking and gave a TEDx-style talk, "The Happiness Paradox."
Laith. Since 2022 I have run a YouTube channel that helps high-school students get into competitive university programs. Watch on YouTube
Have a role or a project? Let's talk.
Tell me what it is and I will reply by email.