Ahmed tarek Metwallli
AI System Engineer
New to Contra
Ahmed tarek is building their profile!
Built an interactive Power BI dashboard for film industry analytics, visualizing key performance metrics including global box office revenue, production budgets, and opening day sales across 150 unique directors. The dashboard features director influence rankings on opening sales, correlation analysis between ratings and budgets, and monthly revenue trend breakdowns to identify seasonal performance peaks.
0
2
This project focuses on building a deep learning classification system for medical signal analysis. A neural network model was developed to classify multi-channel physiological data into two distinct categories (Channel 1 and Channel 2), supporting automated diagnostic decision-making.
The workflow involved preprocessing raw medical signals through standardization and noise reduction, engineering discriminative features from the data, and training a deep neural network with dropout regularization and batch normalization for robust classification. Model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics.
Exploratory data analysis was conducted using pairplot visualizations to assess feature separability between the two channels. The results show moderate to strong class separation across multiple feature combinations, confirming that the extracted features carry meaningful diagnostic information. Kernel density plots further reveal which individual features contribute most to distinguishing between the two classes.
The project was implemented in Python using TensorFlow/Keras, scikit-learn, pandas, NumPy, and Seaborn/Matplotlib for modeling, evaluation, and visualization.
0
3
Zero-Dependency Knowledge OS & Advanced RAG Pipeline
💡 Overview
Most corporate RAG applications rely on a fragile, expensive web of external SaaS vector databases and cloud orchestration APIs that complicate deployment and create data privacy liabilities. This project demonstrates a fully self-contained, 9-feature full-stack AI platform engineered to operate with zero external infrastructure dependencies, delivering enterprise-grade search precision completely within an isolated environment.
🛠️ The Advanced Tech Stack
Orchestration & State Management: LangGraph, Python FastAPI
Vector Database & Storage: pgvector, PostgreSQL
Deployment & Frontend: Docker, React
🧠 Deep-Tech Engineering Milestones
Algorithmic Hallucination Containment: Slashed LLM hallucination rates by ~35% across 200+ academic Q&A pairs.
Advanced Retrieval Mechanics: Moved past naive vector search by implementing a sophisticated semantic RAG pipeline utilizing top-6 vector retrieval and deep cross-encoder reranking running directly on pgvector.
Total Infrastructure Isolation: Eliminated cloud-vendor lock-in by designing a zero-dependency architecture, achieving cloud-ready deployment from a completely cold start in under 2 hours via Docker containerization.
Cyclic Multi-Step Logic: Employed LangGraph to build robust, stateful agentic workflows that keep tracking context across multi-tiered user requests without losing state.
0
6
Project Title: High-Availability Arabic Dialect NLP Pipeline & LLM Router
💡 Overview
Standard LLM implementations fail when processing localized dialects and collapse under heavy concurrent user traffic. This project showcases a high-availability, production-ready NLP pipeline capable of handling complex linguistic nuances while maintaining strict enterprise uptime boundaries.
🛠️ The Tech Stack
LLMs & Frameworks: Llama-3, Mixtral
Vector Database: ChromaDB
Core Logic: Python, Custom Rate-Limiting & Model-Rotation Logic
🚀 What’s Happening in the Video
The attached video demo showcases the end-to-end operational pipeline, demonstrating:
Dialect Resilience: Real-time processing and structural classification of mixed, highly localized dialects.
Failover Execution: Live simulation of concurrent user spikes triggering the custom model-rotation backend to prevent token rate-limit starvation.
📈 Measurable Impact & Engineering Milestones
Multilingual Knowledge Base: Built a specialized RAG pipeline utilizing over 4,500 multilingual embeddings.
Dialect Expansion: Enabled highly accurate few-shot classification across more than 10 distinct Arabic and Egyptian dialects.
System Resilience: Engineered model-rotation and rate-limiting logic that slashed system inference failures by ~40%.
Enterprise-Grade Availability: Sustained a proven 99%+ system uptime under intense, concurrent simulation loads.
📝 Project Role & Context
Role: Lead AI Systems Engineer
Context: Developed as a high-performance solution for the Deep X Hackathon Explainable AI Challenge.
0
10