00 — Profile

Abdelrahman Kerdawy

ML Engineer & Software Developer · Cairo, Egypt

I build things that see, think, and move data — from YOLO models to Flutter apps to RAG pipelines.

01 — About

A working engineer, not a brand.

I'm an engineer in Cairo. Most of what I build lives at the seam where code meets the physical world — a YOLO model running over recorded video, a PIC microcontroller polling a load cell, a Flutter app talking to a Python service on EC2. That seam is where problems get interesting, which is why I went into computer engineering. Nothing about it has gotten boring yet.

I freelance full-time on Upwork while finishing my final year of university. Most weeks I'm shipping something to a paying client and writing exam prep on the same laptop. I'm not chasing buzzwords — I want systems that survive the off-day, the bad input, the dropped network. That bias toward will this hold shows up in the code.

The work below is a sample of what I've shipped along the way. Quietly building toward my own SaaS in the background.

02 — Selected work

8 projects, 8 problems.

D:/Program Files/Git/01

Wall Crack Detection

Computer vision � Inspection

Problem

Inspecting walls for structural cracks by eye is slow and inconsistent, especially across batches of field photos.

Approach

Trained a YOLOv8 detector on annotated crack imagery, then wrapped inference in a Python service that classifies cracks by type and renders annotated output for the inspector.

Stack

YOLOv8 · PyTorch · Roboflow · OpenCV · Python

Result

Built for a real inspection workflow � runs over batches of field photos and outputs annotated images with class and bounding box.

/02

Face Verification System

Deep learning · Identity

Problem ·A client needed to verify that two face images belong to the same person without sending data to a third-party API.

Approach ·An embedding pipeline built on a custom-trained MobileNet variant, exposed through a Python REST API so the verification step can be called from any client app.

PyTorch · MobileNet · FastAPI · OpenCV · Docker

Result ·Deployed as a private API for the client. No data leaves their server.

/03

WhatsApp RAG Chatbot

Conversational AI · RAG

Problem ·A small business wanted customers to ask questions over their product catalog and policy docs through WhatsApp instead of digging through a website.

Approach ·Indexed the documents in ChromaDB, built a LangChain retrieval chain over Gemini, and wired it to the WhatsApp Business API so messages flow in and out automatically.

LangChain · ChromaDB · Gemini · WhatsApp Business API · Python

Result ·Live for the client. Replaces a manual FAQ inbox.

/04

IoT Horse Feeder

Embedded · Hardware ↔ software

Problem

A horse owner needed scheduled, weight-measured feeding when no one's at the stable, plus remote monitoring.

Approach

PIC microcontroller drives the actuator and load-cell loop; a Raspberry Pi handles scheduling, network, and the Python backend. The boards talk over UART; the backend exposes status to a web dashboard.

Stack

PIC microcontroller · Raspberry Pi · Python · UART · load cell · stepper actuator

Result

Working prototype that bridges firmware and software end-to-end — the kind of system most software-only people don't get to build.

/05

AI Instagram Comment Generator

Generative · Marketing

Problem ·Brands wanted contextual, on-tone comments on target posts at scale, without obvious AI tells.

Approach ·A generation pipeline that takes the post's caption and image context, prompts a model under brand-voice constraints, and surfaces options for human review before posting.

Python · OpenAI / Gemini · FastAPI · web UI

Result ·Internal SaaS-style tool. Not yet public.

/06

Breach Scanner

Security · OSINT

Problem ·Users wanted a fast way to check whether their email or domain shows up in known credential leaks.

Approach ·A Python service that queries multiple breach data sources and aggregates the results into a single response, with a small frontend on top.

Python · FastAPI · external breach APIs

Result ·Useful for any team doing basic credential hygiene. Designed for personal and small-team use.

/07

Flutter App + Python Backend

Mobile · End-to-end

Problem ·A client needed a cross-platform mobile app with real backend logic, not a thin wrapper around someone else's API.

Approach ·Mobile UI in Flutter, backend in Python with REST endpoints, both deployed — backend on AWS, app published through normal store channels.

Flutter · Dart · Python · FastAPI · PostgreSQL · AWS EC2

Result ·Shipped to a real client. End-to-end mobile and backend work, not a prototype.

/08

Keyword Density Orchestrator

SEO · LLM tooling

Problem ·SEO writers waste time tuning keyword density by hand across long pieces of content.

Approach ·A small orchestration tool that analyses a draft, computes density against target terms, and uses an LLM to propose sentence-level edits where density is off — without rewriting voice.

Python · OpenAI / Gemini · web UI

Result ·Internal tool. Useful enough that I keep using it on my own writing.

03 — Practice areas

Where I work, and what with.

Computer Vision & ML: Training and shipping detection models on real images, not toy datasets.; YOLO · PyTorch · MobileNet · pipelines
Backend & APIs: Python services that hold up under real traffic, with a real database behind them.; Python · FastAPI · Flask · PostgreSQL · AWS
Mobile Development: Cross-platform Flutter apps that actually reach a store, with a Python backend on the other end.; Flutter · Dart · Python integration
Automation & AI Agents: RAG pipelines, agent orchestration, and chat interfaces that move work off humans.; n8n · LangChain · ChromaDB · Gemini · WhatsApp
IoT & Embedded: The seam between firmware and a Python service — UART, sensors, real signals.; Raspberry Pi · PIC · sensors · HW↔SW bridging
DevOps & Cloud: The infrastructure that keeps the interesting parts running — boring on purpose.; Docker · GitHub Actions · AWS S3 / EC2 / Lambda · Vercel

04 — Get in touch