kahledawad2019-netizen

Pharmacist turned AI builder

I combine healthcare judgment with end-to-end technical delivery. My work emphasizes grounded retrieval, reproducible evaluation, explainability, testing, and systems that people can actually use.

Open to AI Engineer, Machine Learning Engineer, Data Scientist, Data Engineer, Data Analyst, and BI opportunities.

Evidence at a glance

GraphRAG platform	Healthcare RAG	Brain MRI AI	Multimodal ML
21,854 vectors · 1,511 graph nodes	100% drug recall@5 · 95% fresh held-out field recall@5	94.9% test accuracy · 0.948 macro F1	90.07% accuracy · 91.9× median latency reduction

Selected systems

Digiler AI

Adaptive learning platform combining hybrid retrieval, GraphRAG, multimodal search, student modelling, reinforcement learning, and coordinated AI agents.

257 indexed resources and 21,854 vectors
1,511-node / 5,091-edge concept graph
1.00 grounding and citation accuracy across 40 GraphRAG queries
240 passing platform and backend tests

Python GraphRAG Qdrant FastAPI Agents

AdaptiveScan

Explainable brain-MRI classification and localization with EfficientNet-B2, Grad-CAM/Grad-CAM++, independent segmentation, human feedback, and structured reporting.

94.9% test accuracy
0.948 macro F1
0.988 macro AUC
Leakage-aware image split and production operating point

PyTorch TensorFlow FastAPI Streamlit XAI

HYDRA-Net

Cascaded multimodal counter-UAV detection with confidence-gated early exit.

90.07% accuracy and 0.9003 macro F1
0.9812 one-vs-rest ROC-AUC
91.9× median latency reduction
14 passing unit tests

PyTorch Multimodal Learning Signal Processing Early Exit

Steel Defect Detection

Six-class surface-defect detection with a five-seed statistical gate and measured PyTorch, ONNX, and TensorRT deployment benchmarks.

0.7475 ± 0.0161 mAP@0.5 across five seeds
5.73 ms latency / 174.6 FPS on the measured setup
Negative ablation results reported rather than hidden

YOLOv8 PyTorch ONNX TensorRT Statistics

PhARMA RAG

Grounded drug Q&A using dense and lexical retrieval, reranking, citations, refusal safeguards, and deterministic interaction lookup.

506 FDA drug records and 14,955 retrieval chunks
100% drug recall@5
95% recall@5 on a fresh held-out field set
26/26 end-to-end checks documented

Python FAISS BM25 LLM Docker

Hotel Cancellation Prediction

Seven-member end-to-end analytics project spanning SQL, Python EDA, feature engineering, ensemble modelling, and Power BI.

119,208 bookings and 60+ engineered features
92.93% accuracy
0.9495 F1 and 0.9766 ROC-AUC
Eight-model comparison with a stacking ensemble

SQL Server Python XGBoost Power BI Team Project

Capabilities

AI and retrieval: RAG, GraphRAG, hybrid search, reranking, vector databases, grounded generation, evaluation, agentic systems
Machine and deep learning: scikit-learn, XGBoost, PyTorch, TensorFlow, YOLO, forecasting, multimodal learning, model optimization
Data and analytics: Python, SQL/T-SQL, Pandas, NumPy, Power BI, DAX, Power Query, Microsoft Fabric, Excel
Engineering and deployment: FastAPI, Streamlit, REST APIs, Docker, GitHub Actions, testing, ONNX, TensorRT
Domain: pharmacy, healthcare data, clinical context, medication information, safety-oriented communication

Education and credentials

Scholarship Diploma in Data Analytics & Applied AI — Digilians, Egyptian Military Academy in cooperation with Egypt's Ministry of Communications and Information Technology; completed August 2026
Bachelor of Pharmacy — Mansoura University, 2023
Microsoft Certified: Fabric Analytics Engineer Associate
Microsoft Certified: Power BI Data Analyst Associate

Engineering principles

Verify claims against committed evidence.
Report failed experiments and uncertainty honestly.
Treat grounding, refusal, explainability, and data leakage as engineering concerns.
Optimize for reproducibility and practical use—not badge count.

Contributions

Healthcare insight. Production-minded engineering. Measurable impact.
Start a conversation

Navigation Menu