Mohamed HANNANI, LLM Engineer

• Architect and deploy production RAG pipelines using LangChain, Qdrant vector database, and Claude/OpenAI APIs for healthcare document retrieval and intelligent Q&A systems, improving information access efficiency by 40%.
• Engineer healthcare chatbot with Elasticsearch-powered retrieval, sentence-transformers embeddings, and context-aware generation, processing 1000+ daily user queries with 95% user satisfaction.
• Build automated web scraping pipelines with Playwright and BeautifulSoup4 for data extraction and vector database ingestion, reducing manual data entry by 85%.
• Develop geospatial mapping applications using Leaflet, Deck.gl, and PostGIS for hospital network visualization and pharmaceutical territory analysis across 200+ healthcare facilities.
• Design interactive dashboards with D3.js, React Flow, and custom components for real-time healthcare KPI monitoring and executive decision support.
• Manage AWS EC2 infrastructure orchestrating 50+ Docker containers with CI/CD pipelines, health monitoring, and zero-downtime deployments.

• Developed in-context machine translation systems using LLMs (GPT-4, Claude Opus, LLaMA 2) for generic text and subtitle translation, achieving 25% BLEU score improvement over baseline transformer models.

• Led 3-person ML team to deploy state-of-the-art OCR and information extraction models for Sky Onboarding™ platform, reducing document processing time by 30% and improving accuracy by 15% across 156+ document types.
• Deployed scalable ML models for Sky ID™ identity verification using Docker and Kubernetes on AWS, implementing A/B testing framework to optimize model performance and achieving 99.2% uptime.
• Engineered signature detection and extraction system for Sky Signature™ using object detection models (YOLO, Faster R-CNN), processing bank checks and contracts with 97% precision across 62+ check formats.
• Built production OCR pipeline with Flask API and Tesseract/PaddleOCR integration for bank check validation, achieving 95% accuracy and eliminating manual verification for 10,000+ monthly transactions.
• Developed automated information extraction system using transformer models and custom NER pipelines for bill processing, achieving 98% accuracy and improving data processing efficiency by 70%.

Empfio — AI Appointment Booking SaaS

Production AI SaaS handling real customer conversations and booking appointments 24/7 across WhatsApp, SMS, web chat, and voice — solo-built from zero to launch. Architected 6 interconnected services: FastAPI backend (15+ DDD domains, async PostgreSQL, Redis, Celery), LangGraph AI agent with multi-LLM support (GPT-4o, Claude), real-time voice pipeline with ~500ms response time, Next.js 14 dashboard (8 languages), and Stripe billing with API-driven feature registry.

ECO Analyzer – AI-Powered Healthcare Analytics Platform

Designed and built a scalable healthcare analytics platform for German ASV (Ambulant Specialist Care) data, enabling data-driven insights for healthcare administrators and researchers. Developed a microservices architecture using FastAPI, Node.js, PostgreSQL + PostGIS, and Redis to analyze 18,000+ doctors and 200+ specialist care teams across Germany. Implemented geospatial analytics, healthcare network collaboration mapping, and AI-powered insights (GPT-4 / Claude) through interactive dashboards built with Next.js, D3.js, and Leaflet. Focused on performance, security, and enterprise-grade scalability in a regulated healthcare environment.

Deutsch Tutor - AI-Powered German Vocabulary Learning Bot

Built an AI-powered WhatsApp learning system applying structured morpheme-based word decomposition (Wortzerlegung) for vocabulary acquisition. Developed an async FastAPI backend with PostgreSQL (SQLAlchemy 2.0 async), OpenClaw gateway integration, and Claude AI for real-time linguistic analysis. Implemented SM-2 spaced repetition, quiz workflows, personalized vocabulary tracking, and webhook-based message routing. Containerized with Docker and managed schema evolution via Alembic, ensuring clean architecture and scalable async processing.