Mohamed HANNANI, AI Engineer

• Built and operationalized LangChain agent pipelines backed by a Qdrant vector store across 200+ healthcare facilities — document retrieval, Q&A, and structured extraction — cutting manual processing effort by 40%.
• Defined reusable patterns for LLM integration (prompt design, RAG, structured extraction, model routing) and made principled model-selection calls based on cost-performance trade-offs and inference latency.
• Built interactive KPI dashboards (D3.js, React) enabling non-technical stakeholders to self-serve operational insights, reducing ad-hoc reporting requests by 60%.
• Provisioned AWS infrastructure (EC2, Lambda, S3) with Terraform and CI/CD, orchestrating 50+ Docker containers with zero-downtime deployments and the safeguards required in a regulated industry.
• Identified and prioritized AI opportunities end-to-end — assessed business impact, proposed solutions to stakeholders, and shipped them into daily use.

• Built LLM reasoning pipelines (GPT-4, Claude, LLaMA 2) for automated in-context machine translation — prompt engineering, chain-of-thought, retrieval-based context — 25% BLEU improvement over baseline.
• Designed evaluation frameworks to compare prompts, contexts, and models systematically — turning subjective LLM choices into measurable decisions.

• Led a 3-person team building modular, reusable ML inference services across 156+ document types — component reusability reduced integration effort by 30% per new format.
• Architected RESTful API services (Flask) linking ML inference systems to downstream business workflows — 10,000+ monthly transactions, structured logging, 99.2% uptime. Deployed on AWS with Docker and Kubernetes.

Empfio — Production AI Agent Platform (Solo-built)

Solo-built in ~3 months: a production AI agent platform that runs the front desk for service businesses. Autonomous agents across voice, WhatsApp, web chat, Telegram, and SMS — intake, intent classification, appointment booking, customer payments (Stripe Connect), follow-ups, and escalation to a human when needed. Multi-LLM routing (GPT-4o, Claude, Groq, Ollama) with per-org circuit breakers, MCP-shaped tool registry, RAG with vector retrieval, and topic-driven behavior configuration (new vertical = configuration, not new code). Full observability via Langfuse, Prometheus, and OpenTelemetry GenAI semantic conventions. Built on FastAPI async services, PostgreSQL, Redis, Celery, LiveKit, Next.js.

ECO Analyzer — Healthcare Intelligence Platform

Built a healthcare analytics platform over German Ambulant Specialist Care (ASV) data — 18,977+ doctors, 247 specialist teams, 688 cities. Interactive geospatial map (Leaflet, PostGIS) with network analysis, plus an AI Copilot (GPT-4) that lets users query the dataset in natural language — a shared retrieval pattern reusable across teams and use cases. 8 independently deployable microservices (FastAPI, Node.js) over PostgreSQL + PostGIS and Redis.