AI Engineer · Healthcare Tech · Enterprise Platforms
📍 Damascus, Syria · 📞 +963 997 300 716 · ✉️ mohamad.ashmar.ai@gmail.com
🔗 GitHub · quantum-ai
Software Engineer (DTU, Denmark) with 5+ years of experience spanning the full AI lifecycle — from training models on hospital data and deploying on SuperComputer-DTU, to building hardware-software medical devices from scratch, to architecting multi-tenant enterprise platforms serving real users at scale. I've secured $2M+ in funding for a Danish Board of Health-approved data extraction project, delivered 15+ production AI modules on GKE with ArgoCD GitOps, and driven measurable client impact including 60% profitability gains and 35% cost reduction. My work covers healthcare diagnostics (kidney/bladder tumor detection at 85–98% accuracy, real-time 3D urological devices), enterprise SaaS (14 analytics modules with SARIMA/Prophet forecasting, 10+ FastAPI microservices with circuit breakers and Redis Vector), LLM/RAG systems (GraphRAG, multi-agent architectures, 11+ MCP servers), embedded systems (PCBA validation, IoT-Azure integration), and governmental AI strategy (advising the Damascus Health Directorate on national AI adoption). I navigate complex regulatory landscapes including MDR 2017/745, ISO 13485, HIPAA, and GDPR, and have led cross-functional teams across Denmark, Saudi Arabia, and Syria.
| Metric | Achievement |
|---|---|
| $2M+ | Funding secured for Epic healthcare data extraction (Danish Board of Health approved) |
| 90% | Benign vs malignant kidney tumor classification |
| 98% | Kidney vs non-kidney detection |
| 94% | Four-class kidney classification (benign/malignant/normal/undefined) |
| 90%+ | Bladder tumor detection — suspicious region detection in live video from cystoscope, deployed on Onyx device for real-time diagnosis |
| 60% | Client profitability increase via AI-driven inventory optimization |
| 35% | Operational cost reduction |
| 15+ | Production AI modules on GKE (FastAPI + ArgoCD) |
| 11+ | MCP servers in production |
| 3 | Countries of cross-functional team leadership (DK, KSA, SY) |
AI/ML: PyTorch · TensorFlow · JAX · HuggingFace Transformers · Unsloth · TRL · Axolotl · DeepSpeed
RAG & Agents: LangChain · LlamaIndex · LangGraph · GraphRAG · CrewAI · AutoGPT
CV & Multimodal: OpenCV · YOLO · SAM · Detectron2 · Diffusers
Infrastructure: FastAPI · GKE · ArgoCD · Docker · Terraform · Prometheus · Grafana
Databases: Elasticsearch · Redis · PostgreSQL · Neo4j · Qdrant · Pinecone · Milvus
LLM Serving: vLLM · TGI · ONNX · TensorRT · OpenRouter · Groq
Languages: Python · Rust · Go · TypeScript · Java · C++
Spearheaded a 1.5-year collaborative project between RUC and DTU using RPA to extract and stabilize patient data from Epic Software. Modular Python architecture (PyAutoGUI, OpenCV, pytesseract) for screen scraping and OCR — approved by the Danish Board of Health. Automated upload to REDCap within HIPAA/GDPR/21 CFR Part 11 compliance. Secured $2M+ in funding to expand across Europe.
Pipeline using GPT-4 and open-source LLMs (LLaMA) to process patient data extracted from healthcare systems. Generated clean structured Excel outputs, predicted patient risk percentages pre/post-operation, implemented RAG for local database Q&A, and built anonymization software (static + dynamic algorithms) for patient privacy before sending data to external models.
📦 hospital-data-engineering-RaD
Four AI models deployed on DTU SuperComputer: (1) Kidney vs. Non-Kidney detection (98%), (2) Benign vs Malignant classification (90%), (3) Four-class classification — benign, malignant, normal, undefined (94%), (4) Localization/Segmentation (YOLOv8) + Tumor Detection. Automated Jenkins/Docker CI/CD pipeline (RCC_v2) integrating with Orthanc PACS via C-STORE SCU protocols. Multi-stage processing: Data Validation → DICOM to BMP → AI Classification → Cleaning → Segmentation. SNOMED/ICD code mapping from raw Excel exports for oncology metrics (TNM staging, ISUP grading).
📺 Project Demo · 📐 System Design · 📄 Docs · 📦 DICOM Toolchain · Hospital UI
Real-time AI system detecting bladder tumors from cystoscopic video footage using SSIM (Structural Similarity Index) for intelligent sub-sampling and KMeans Clustering for key frame extraction — eliminating redundancy while enabling urologists with actionable intra-procedure insights.
📦 Bladder Model · Video Extractor
Built a medical device from scratch integrating Ambu® aScope™4 Cysto with BNO055 orientation sensor for real-time cystoscope tracking mapped onto a 3D bladder model with color-coded navigation. SmartScopePC (AMD Ryzen 9 7950X, RTX 4090) for real-time inference. FastAPI microservice backend with client-server architecture enabling remote processing on DTU Supercomputer. ISO 13485 / MDR 2017/745 compliant.
📺 Project Demo · 📄 Tracking Docs · Hardware Specs · 📦 Stream Service · Stream Client · Web Integration
Touchless interface enabling surgeons to control operational equipment via hand gestures in the air, maintaining sterile field. Computer Vision pipeline (MediaPipe + OpenCV) for full hand/finger detection and gesture recognition. Kinematic logic using np.arctan2 and Euclidean distance for sub-millimeter precision. Hardware Abstraction Layer (HAL) parsing IMU telemetry with BLE→WiFi 6 communication roadmap for zero-lag response.
End-to-end PCBA process management with Eurocircuits (ODB++, CPL, BOM). AI-driven PCB design ecosystem using LLMs for component placement and routing optimization. Embedded firmware flashing (Atmel-ICE, Microchip Studio). LoRa→Azure Cloud IoT integration with real-time data visualization. Three-phase power system fault simulation for AI model testing.
📄 Hardware Docs · PCBA Docs · Firmware Docs · 📦 AI Designer · PCB Design
Class II Medical Device compliance (MDR 2017/745, ISO 13485, ISO 14971, FDA 21 CFR Part 820). Collaborated with Presafe Denmark A/S and Dansk Standard for CE Marking. AI integration strategy under EU Artificial Intelligence Act. Cost-benefit analysis for regulatory approval ($70K–$140K) and eQMS implementation.
Founded VisionTechWave in Denmark, driving AI and healthcare innovation. Directed RCC AI Intelligent Decision Support System for radiologists (Agile Sprint planning, full lifecycle from design to deployment).
📄 Business Docs · 📦 XRay Repo
Distinguished between house types from user-provided images using Computer Vision (SAM, Stable Diffusion). Later expanded to exterior facade design — detecting and replacing windows while preserving original view. Achieved the highest score in the Danish system (12/12). Demo · Thesis
Embedded device communication software in C++ (Linux-64bit/Windows-64bit), DLL libraries, TR31 secure key interchange, CMake, GDB debugging, Doxygen documentation, Jenkins CI/CD with Groovy. Developed mock testing framework for NUC management with pytest integration into Jenkins pipeline.
Side projects during this period: 📦 ambassador-microservices-golang (Golang, Microservices, K8s, Stripe) · KodeKloud Engineer · 📺 Demo
quantum-ai-web.netlify.app — "Building the Future with AI كوانتم" — We design and deploy cutting-edge AI solutions that transform businesses, automate operations, and unlock unprecedented growth opportunities. GitHub
Multi-tenant FastAPI platform on GKE with SARIMA/Prophet forecasting, menu engineering (BCG Matrix), inventory optimization (98% accuracy), and organization-scoped Elasticsearch isolation. Enabled 60% profitability gains and 35% cost reduction for clients.
FastAPI microservices monorepo with 17+ independently deployable services on GKE via ArgoCD GitOps. Includes OCR invoice extraction (4-stage pipeline, 50× cost savings), PO-Invoice Matcher (3-phase financial reconciliation with GRN auto-suggestions), AI Insights Chatbot (53 business modules, WebSocket + HTTP, Redis Vector for 18–52× faster search), Text-to-Image (3-agent LLM pipeline), and AI Recipe Generation (Halal compliance, bilingual).
Suite of production Model Context Protocol servers including Wateen operations (5 servers, 8,000+ LOC), Snowflake administration (55+ tools, 7,800+ LOC), and enterprise integrations (Elasticsearch, Keycloak, GCloud, LinkedIn Marketing, Browser automation).
ChatDev (virtual software company with role-playing agents), Reachware AI Agents (production assistant, code generator, proposal evaluator with TinyLlama 1.1B 4-bit + QLoRA), 6-service food microservices (CrewAI + ChromaDB + Apache Pulsar), Anything-LLM full-stack RAG (27+ AI providers, 8 vector DBs, MCP protocol).
12 specialized AI agents (EstimationBot, ContractReviewAI, BOQ generation from PDF/DWG/IFC/BIM) using FastAPI, Qdrant, MCP Protocol, and A2A agent interoperability. AI-powered quantity takeoff with CSI MasterFormat codes and historical pricing.
Strategic advisor to the Directorate Head: AI readiness audits across public hospitals (Damascus Hospital, Al-Mowasat Hospital), ministry-level presentations on national AI adoption roadmap. Deployed bilingual (Arabic/English) administrative dashboard for 200+ employees across 22 facilities with employee management, project oversight, real-time Plotly analytics, and operational alerts.
📄 Ministry AI Strategy Presentation · 🌐 Dashboard
Designed and delivered comprehensive AI curriculum. Trained 11 students to full AI engineering proficiency. Mentored 2 high school students to build a 90%+ accurate heart disease detection model. Founded mentorship program at Damascus University (50+ attendees per workshop). Workshops on Agentic AI, System Prompt Engineering, Full-Stack (React+AI), Back-End, and Mobile Development (Flutter & React Native).
📺 Course Instagram · Workshop Instagram
- LLM-From-Zero-to-Hero — Comprehensive hands-on curriculum for mastering LLMs from absolute beginner to production practitioner
"Building AI that ships — from hospital wards to Kubernetes clusters."