class DataEngineer:
def __init__(self):
self.name = "Karim Magdy"
self.role = "Data & AI Engineer"
self.university = "Cairo University 🎓"
self.language_spoken = ["ar_EG", "en_US"]
def say_hi(self):
print("""
Thanks for dropping by! I'm passionate about
bridging raw data and actionable intelligence.
Let's build something amazing together! 🚀
""")
me = DataEngineer()
me.say_hi()- 🔄 Transform raw data into insights using ETL workflows & visual analytics
- 🐍 Leverage Python, SQL, and distributed systems for complex data challenges
- 🧠 Specialize in Deep RL, NLP, and Computer Vision
- 🤖 Build GenAI applications with LangChain, LangGraph & LLMs
- 📊 Design scalable data pipelines and production ML/MLOps systems
- 🛠️ Engineer robust backend systems with Node.js, FastAPI & .NET
🏢 Click to expand my experience
📅 Jul 2025 – Sep 2025 | 📍 Giza, Egypt
| 🔧 | System Modernization: Re-architected Fleet Management database using .NET technologies |
| 📦 | Data Warehouse: Built centralized warehouse with automated ETL pipelines |
| 📊 | Analytics: Developed interactive Power BI dashboards for executive decision-making |
📅 Aug 2024 – Sep 2024 | 📍 Cairo, Egypt
| 👑 | Leadership: Served as Team Leader, guiding technical solutions research |
| 📡 | Big Data Analysis: Processed large-scale 5G datasets using Python |
| 🔍 | Visualization: Designed drill-down filters in Power BI for fault analysis |
A self-hosted, Kubernetes-native Zero-Trust platform that closes the loop between detecting anomalous user behavior and enforcing access — an open-source, affordable alternative to commercial SASE for SMEs and sovereignty-constrained organizations. Instead of trusting an identity once at login, it maintains a continuous, per-user trust score that falls as behavior drifts from a learned baseline and acts on it automatically, with no human in the request hot path.
| 🧠 | Stateful behavioral AI: an LSTM-autoencoder with per-user identity embeddings scores a real-time UEBA stream — Fluent Bit → Kafka → Apache Flink (5-min sliding / 1-min slide windows) → 28-dimensional per-user behavior vectors. Sequence state is persisted in Redis + PostgreSQL for crash recovery and zero-downtime model hot-swap. |
| ⚖️ | Closed-loop enforcement: a Policy Engine folds noisy anomaly scores into a bounded cumulative trust score (EWMA, threshold-aware penalties, escalation), and a Trust Actuator enforces out of band — step-up MFA via Keycloak or an edge block-list — over an Istio mTLS mesh with deny-by-default authorization. |
| 🔁 | Continuous MLOps: champion/challenger retraining gated on a fixed golden set, auto-labeled from the system's own enforcement decisions, with live model promotion from the Admin Portal — which also ships an on-cluster-LLM SOC Copilot, a manual-override Penalty Box, and an attack-simulation sandbox. |
|
|
|
|
|
|
|
|
|
-grey?style=flat-square){kind=icon}
-blue?style=flat-square){kind=icon}
I'm always open to interesting conversations and collaboration opportunities.
