Projects
A showcase of my technical skills, problem-solving abilities, and passion for creating impactful solutions
History of Present Illness Generator
AI-powered tool for automating extraction and structuring of patient data for radiation oncology workflows
Impact & Achievements
- Reduced manual compilation time from 40+ minutes to 30 seconds
- Pioneered AI-integrated workflows for cancer cases
- Enhanced efficiency in radiation oncology patient care
Technologies Used
AI Agent-Driven Data Visualization Platform
A platform featuring autonomous AI agents that dynamically switch between data sources, execute complex queries, and generate real-time interactive dashboards using ECharts visualization engine through Code Tooling
Impact & Achievements
- Automated complex data visualization query execution
- Reduced dashboard creation time from hours to minutes through AI automation
- Implemented intelligent query caching reducing response times by 80%
- Advanced predictive analytics with machine learning-powered insights
Technologies Used
Generative Editing Feature for Patient Notes
LLM-powered feature for physicians to refine and enrich patient notes with AI-generated suggestions and revisions
Impact & Achievements
- Enhanced quality and detail of medical documentation
- Optimized efficiency and accuracy in healthcare settings
- Automated comprehensive revision suggestions
Technologies Used
Clinical Notes De/Anonymizer
HIPAA-compliant Python tool for replacing Protected Health Information (PHI) in medical documents using advanced NLP models
Impact & Achievements
- Achieved 99.3% recall across 50k+ documents
- Deployed on dedicated GPU VM for secure PHI handling
- HIPAA-compliant anonymization workflow
Technologies Used
Radiotherapy Workflows RAG Assistant
Retrieval-Augmented Generation system for automating clinical protocol retrieval and generating context-aware responses
Impact & Achievements
- Achieved sub-second latency (<1s)
- Improved clinician response quality significantly
- Automated retrieval of radiotherapy guidelines
Technologies Used
Data Synthesizer for Network Security Models
Creative data-driven model for synthesizing time-series datasets to improve network security model accuracy and fidelity
Impact & Achievements
- Achieved State-of-The-Art 81.1% accuracy (+5pp improvement)
- Outperformed DoppelGANger paper
- Trained on 100k data points using distributed GPUs
Technologies Used
YouTube Videos Stance Classifier
Machine learning model that predicts YouTube video stances on controversial topics by analyzing user comments using NLP
Impact & Achievements
- Achieved 80.3% accuracy across five controversial topics
- Implemented three ML algorithms for comparison
- Analyzed large-scale YouTube comment datasets
Technologies Used
Speech Captioner for English Videos
Automatic subtitle generation system using Wav2vec 2.0 Transformer for English video content with precise timing alignment
Impact & Achievements
- Achieved 1.8/3.3 word error rate
- Precise text-to-video frame alignment
- Real-time subtitle generation capability
Technologies Used
Font Matching Generator - OCR
OCR model for identifying and classifying fonts in old Arabic books using computer vision techniques
Impact & Achievements
- Specialized in historical Arabic text recognition
- Built during internship at RDI-EG
- Advanced font classification capabilities
Technologies Used
Neural Machine Translation
Seq-to-seq translation model implementation using GRUs and Attention Mechanism built with PyTorch
Impact & Achievements
- Custom implementation of attention-based translation
- Comprehensive seq-to-seq architecture
- Educational implementation of modern NMT techniques
Technologies Used
Appointment Volume Predictor
Predictive analytics system for healthcare appointment scheduling optimization to reduce staffing costs
Impact & Achievements
- Reduced staffing costs by 12%
- Optimized appointment scheduling workflows
- Improved resource allocation efficiency
Technologies Used
Prompt Engineering & RAG Pipeline
End-to-end Retrieval-Augmented Generation workflow with vector stores and embedding optimization for domain-aware LLM responses
Impact & Achievements
- Achieved sub-second latency (<1s)
- Built comprehensive RAG infrastructure
- Optimized embedding and retrieval performance