- Key Engineer across three Oil & Gas engagements spanning the Middle East and Europe — architecting AI multi-agent systems that automated domain-specific business workflows for enterprise clients.
- Adopted Spec-driven Development, shipping production-quality MVPs in days and cutting typical delivery timelines by 75%+.
- Built agentic systems to drive complex, domain-specific business logic end-to-end — from natural-language intake through tool orchestration to structured output delivery.
Hi, I'm Osama.
Senior Data Scientist · AI Engineer
I build agentic systems that turn data into decisions.
Over 8+ years I've gone from training CNNs on ID-card numerals to architecting multi-agent systems that serve 300 people in real time. I care about one thing: turning messy operational data into shipped systems that move real numbers.
- 0Years in DS / ML
- 0Agentic systems shipped to enterprise
- 0Concurrent users served in production
- 0Faster MVP delivery via spec-driven dev
Where I've made an impact
Selected outcomes across consulting, telecom, research, and applied ML — expand any role to see how the work shipped.
Lead Data Scientist
Riyadh, Saudi Arabia · Jan 2024 — Aug 2025- Defined and led the company's GenAI strategy through research, vendor engagement, and hands-on prototyping — establishing the tech stack, use cases, and measurable business value.
- Designed and productionized a streaming multi-agent chatbot assistant using LlamaIndex, Redis, Elasticsearch, FastAPI, and Azure OpenAI — serving 300 concurrent users with RBAC, custom LLM safeguards, and live monitoring via Arize Phoenix.
- Led the digitization of 5TB of scanned documents by architecting a pipeline using Azure Document Intelligence, KNN clustering, and LLMs — indexing structured output in Elasticsearch for semantic search.
- Led deployment of containerized deep learning CV models as ONNX microservices, now used daily by 200+ field technicians.
- Introduced a standardized CRISP-DM AI development lifecycle and implemented MLOps best practices, reducing inference times from minutes to milliseconds.
- Engaged with Gartner and Accenture to refine GenAI tech stack, assess AI maturity, and provide strategic recommendations to leadership.
Data Scientist
Riyadh, Saudi Arabia · Sep 2021 — Dec 2023Contracted via Masterworks, Norconsult Telematics, and Solutions by STC.
Solutions by STC
Riyadh, SA · Aug 2022 — Dec 2023- Achieved a 27% greater reduction in tickets from the Tower Management System by using ML to mine associations in historical alarm data, surpassing the system's existing rules.
- Discovered opportunities for a 12% revenue increase by clustering similar products in CRM and Billing systems.
- Designed a linear programming AI solution that cut technician assignment time from minutes to seconds.
- Partnered with infrastructure to overhaul the AI architecture, enabling a complete MLOps lifecycle.
- Developed a PoC QA chatbot answering questions over the company knowledge base using LLMs, LangChain, and a vector database.
- Deployed a navigational chatbot to production using Botpress and Kubernetes (OKD), improving website navigation.
Norconsult Telematics
Remote · Sep 2021 — Jul 2022- Led the migration of BI dashboards from Tableau to Power BI, consolidating fragmented reporting into a unified platform.
- Automated ETL pipelines using Python and Apache Airflow, reducing manual data prep time by 60%.
- Developed a predictive maintenance system with LightGBM, achieving up to 72% outage detection accuracy.
Masterworks
Riyadh, SA · Sep 2021 — Aug 2022- Managed Python-based ETL jobs for data extraction and transformation, ensuring smooth integration with BI tools like Qlik Sense.
- Contributed to 3 impactful research projects, resulting in 4 conference papers — 2 as first author.
- Conducted research on human activity recognition using IMUs (inertial measurement units), advancing understanding of movement patterns.
- Built an indoor workout virtual coach combining human pose estimation with AI-driven guidance and personalization.
- Executed computer vision tasks, achieving 98% accuracy training convolutional neural networks to identify Arabic/Hindi numerals in Egyptian ID cards.
- Contributed to textile defect classification using a combination of traditional and AI-based computer vision methods.
// about
Data into value, end to end
I'm a Senior Data Scientist and AI Engineer currently at EPAM Systems, where I design and ship multi-agent systems for enterprise clients across the Middle East and Europe. I'm also a Claude Certified Architect (Foundations) early adopter, which reflects the same hands-on depth I bring to agentic architecture, tool design, MCP workflows, Claude Code configuration, prompt engineering, and context management. Before that I spent four years at TAWAL building the company's GenAI stack from scratch — from digitizing 5 TB of scanned documents to deploying a streaming multi-agent assistant used by 300 people daily. My work spans the full arc: from publishing research on gait recognition to deploying containerized ONNX microservices for field technicians in the field.
-
GenAI & Agents
Architected a multi-agent RAG system for 300 daily users: Claude + LlamaIndex + Azure OpenAI, RBAC, live monitoring.
-
Applied ML
27% ticket reduction. 72% outage detection. Models trained on telecom tower data and shipped to production.
-
MLOps at Scale
From notebook to containerized ONNX microservice: full CI/CD and Databricks deployment, end to end in production.
Testimonials
What the work feels like on the other side of it.
Notes from teammates, clients, mentees, and students — kept concise so they read as signal, not as a testimonial wall.
// projects
Selected work
GenAI · Desktop App
01
SilverBrain
Dump everything in your head, let AI extract the tasks, sort them on an Eisenhower Matrix, and execute with Pomodoro focus — all in one calm desktop app.
- AI Task Extraction
- Eisenhower Matrix
- Pomodoro Focus
- Desktop App
GenAI · Agent Workflow
02
Resume Screening & Filtration Multi-Agent
A multi-agent system powered by LlamaIndex AgentWorkflow and DeepSeek V3 that automatically parses batch resume PDFs, maps qualifications against a job description, scores alignment, and ranks the top-K applicants.
- LlamaIndex
- DeepSeek V3
- AgentWorkflow
- PDF Parsing
NLP · Deep Learning
03
Arabic Text Diacritization (RoBERTa)
Fine-tuned a RoBERTa model on the Tashkeelah corpus to restore missing diacritics in Arabic text. Deployed and trained using PyTorch on dual NVIDIA A100 GPUs via Runpod for accelerated training.
- RoBERTa
- HuggingFace
- PyTorch
- Runpod
- NVIDIA A100
GenAI · Agentic RAG
04
Employee HR Assistant Agent
Built an HR companion that parses corporate policies, queries the company's SQL HR database, drafts and sends emails, and directly interacts with embedded web applications to fill forms and redirect users.
- LlamaIndex
- FastAPI
- SQLAlchemy
- Chainlit
- Agentic RAG
Enterprise Tool
GenAI · Production RAG
05
TAWAL Website Chatbot
Designed and deployed a production chatbot for TAWAL's corporate website. Integrates Elasticsearch for vector search and includes guardrails to screen and reject queries that are irrelevant or out-of-scope.
- LlamaIndex
- Elasticsearch
- Botpress
- Docker
- Kubernetes OKD
Production System
GenAI · Open Source
06
LlamaIndex Core Contribution
Modified LlamaIndex's NLSQLRetriever core to support row retrieval for enhanced database querying context. Patched AgentWorkflow to streamline tool selection loops and integrated role-based tool authorization.
- LlamaIndex Core
- AgentWorkflow
- Text2SQL
- RBAC
- Python
Computer Vision · MLOps
07
Field Image Validation Services
Co-developed and containerized dual deep learning CV models to validate site maintenance images submitted by field technicians. Optimized performance utilizing ONNX Runtime, drastically reducing latency to milliseconds.
- Deep Learning
- Computer Vision
- ONNX Runtime
- Docker
- FastAPI
200+ Daily Techs
Deep Learning · Research
08
Inertial Gait-Based Person Authentication
A Siamese-network biometric framework using accelerometer and gyroscope wearables. Authenticates new subjects at test-time without requiring model retraining. Published and presented at IJCNN 2021.
- Siamese Networks
- TensorFlow
- IMU Signals
- Biometrics
- IJCNN
NLP · Analytics
09
Data Scientist In-Demand Skills Dashboard
Scraped job postings from LinkedIn, extracted skills with a fine-tuned HuggingFace NER model, and built an interactive Streamlit and Tableau dashboard to filter market demands.
- HuggingFace
- Streamlit
- Tableau
- Web Scraping
- NLP
Finance · Islamic Law
10
Zakat Calculator · حاسبة الزكاة
Calculate your overdue Zakat in a visual Gantt timeline. Supports Hijri calendar integration, gold/silver valuation, and features an offline-first responsive design.
- Hijri Calendar
- Gantt Timeline
- Offline-first
- CSS Grid
Dev Tools · Binary
11
ELF File Visualizer
Drag and drop an ELF binary to inspect its structures in real time. Features a section minimap, color-coded hex dump, and full binary header breakdown.
- Binary Parsing
- Minimap View
- Hex Viewer
- Vanilla JS
Deep Learning · IoT
12
Wearable IMU Activity Recognition
Collected multi-channel gait data using 8 simultaneous on-body IMUs from 20+ subjects. Trained machine and deep learning systems to authenticate and classify gait activity, hitting 99.8% test accuracy.
- Deep Learning
- IMU Sensors
- Signal Processing
- Activity Classification
99.8% Test Accuracy
Computer Vision · IoT
13
VCoach: Pose-Estimation Coaching
Investigated and integrated 2D/3D human pose estimation (HPE) with on-body IMU sensors. Designed algorithms to evaluate exercise execution quality and posture dynamically during workouts.
- Pose Estimation
- Computer Vision
- IMU Sensors
- Sensor Fusion
Biofeedback AI
Signal Processing · ML
14
MyoSpeech Signal Classification
Preprocessed and classified high-frequency muscle sensor data (MyoSpeech signals) from speech-impaired patients. Utilized scipy for spectral analysis and scikit-learn for utterance recognition.
- scikit-learn
- scipy
- Signal Processing
- Biomedical ML
HealthTech PoC
// skills
The toolkit
The tools I reach for in production, with visible depth by skill — no hover hunting required.
// certificates
Verified credentials
Supporting proof across data management, cloud analytics, professional development, and applied data science.
A compact record of the credentials behind the work — each one links to the live verification page.
Claude Certified Architect (Foundations)
Verified Anthropic credential covering agentic architecture, tool and MCP workflows, Claude Code setup, and deliberate prompt and context management.
View credential
AWS Certified Data Analytics — Specialty
Cloud credential covering analytics architecture, data pipelines, storage, and AWS-native processing.
View credential
CDMP Associate
Data management credential covering core governance, quality, and information management foundations.
View credential
Applied Data Science Lab
Applied program in model-building, experimentation, and production-oriented data-science workflows.
View credential
McKinsey Forward
Professional development program focused on practical leadership, communication, and problem-solving.
View credential// education
Education
M.Sc. — Egypt-Japan University of Science & Technology
- Graduated with a distinguished GPA of 3.85.
- Awarded a full teaching-assistant scholarship.
- Coursework: Advanced Intelligent Systems, Advanced ML, Learning Algorithms, Neural Networks.
B.Sc. — Faculty of Engineering, Mansoura University
- Graduated Very Good with Honors.
- Major in Computer Engineering & Systems.
// contact
Let's build something intelligent
Open to conversations about data science, applied AI, and multi-agent systems.