Yann Combarnous

Built and led the team that took CBRE’s first AI-native smart-building application from whiteboard to production for a global real estate portfolio to support on-site experience teams, while helping write the AI governance CBRE didn’t yet have. Responsibilities: - Global Team Management: Led and grew distributed engineering and data teams (30+ people) across Europe, North America, and Asia, building an AI-first engineering culture. Handled hiring, capacity planning, hiring-vs-consulting decisions, performance reviews and goal setting. - Budget & Vendors: Ran yearly budget planning and quarterly forecasting for a >€5M/yr engineering budget over five years (OpEx/CapEx, headcount, cloud/infrastructure). Selected and managed delivery vendors, contracts run jointly with Procurement. - Generative AI in Production: Architected and deployed agent-based scheduling optimization, hybrid RAG search (graph and vector), and agentic workflows with human-in-the-loop governance. - Platform & Ecosystem: Delivered a resilient data-broker aggregating multi-partner streams for real-time operational visibility of building signals and AI-driven actions, plus an AI-assisted partner marketplace supporting webhook/queue/API/email integrations and streamed Kafka transformation via low-code DSL. - AI Governance & Influence: Key contributor to enterprise AI governance and security (EU AI Act, GDPR) as the first native AI project, with global Security and Architecture. Led engineering costing for the AI investment case built with Product. Delivered:  - 30 min/day saved per user - Sub-second data latency at millions of records/day - Partner onboarding cut from two weeks to three day - 80% cut in projected annual AI run-cost (~€250k to ~€50k) via model selection, dynamic reasoning-effort optimization, and runaway-loop counter-governors, measured on Datadog LLM Observability - 3x engineering productivity through spec-driven design with agentic coding

Owned the architecture of CBRE’s multi-channel workplace platform for client office workers (showcased on stage by Microsoft's CEO) and its real-time smart-building digital twin, while mentoring the senior engineers around it. Responsibilities: - Cloud-Native Architecture: Architected and implemented a highly available, multi-region, multi-channel platform (mobile, desktop, web) on Kubernetes, with full observability and fault tolerance. - Real-Time & IoT Pipelines: Designed IoT and edge integrations on event-driven architecture (Kafka) and serverless for real-time building data streams, including on-site sensor installation and placement optimization. Delivered real-time indoor navigation using beacons, in-memory caches, and live streams based on a digital twin. - Platform Delivery & MLOps: Built a configuration-driven enterprise SaaS platform with Azure/Google/Okta authorization, SaaS provider integrations (resource booking, visitor, lockers, door systems, occupancy), and deployed building occupancy prediction models to dynamically adjust services. Mentored senior engineers and held technical ownership of key architectures. Delivered:  - Platform serving 100k+ monthly active users across 600+ client offices, with multi-region cloud and failover regions - Mobile/web/desktop application for our clients' office workers, with Intune MDM for security

The innovation function’s technical hands: turning emerging tech into prototypes the business could evaluate and adopt. Responsibilities: - Asset Scanning and Survey: Designed and built “Asset Intelligence” mobile solutions using microservices, NoSQL, and robust message-driven architectures. - Data Ingestion Engine: Designed and built a Data Vault 2.0 engine with a no-code interface for data integration. - Computer Vision R&D: Evaluated and integrated emerging technologies: computer vision, office 3D scanning, conversational AI. - AI Prototyping for Work Order Classification: Prototyped predictive analytics and NLP solutions to extract value from unstructured enterprise data. Delivered: - Accelerated by 2x the process of scanning building assets on site. - Reduced operational data ingestion and transformation development by 50%