Yassir Bourdoukh

• Founded OpsUnit, a consultancy specializing in designing and deploying end-to-end, air-gapped AI infrastructure for highly regulated enterprise environments. • Developed "Loci," an internal knowledge intelligence layer that utilizes RAG (Retrieval-Augmented Generation) to connect operational documentation to a natural language interface, ensuring 100% on-premise data sovereignty. • Engineered a 100% local, air-gapped Data Analysis Agent that enables non-technical stakeholders to perform exploratory "Quick Queries" (NL→SQL) and generate executive-grade reports via a 6-step autonomous pipeline (Schema introspection → Auto-correcting SQL → Statistical anomaly detection → Synthesis), ensuring sensitive corporate data never leaves the internal network. • Architected GitOps-driven deployment pipelines for AI models using GitLab CI and ArgoCD, ensuring predictable, auditable, and version-controlled model governance. • Designed zero-trust security patterns for AI workloads, integrating FreeIPA for identity management, Vault for PKI, and VLAN segmentation to prevent data leakage.

• Led the migration of diverse on-premises solutions to Azure. • Defined cloud architecture for various solutions and managed them with Terraform • Managed ETL processes using Azure Data Factory, ensuring seamless data integration and transformation for improved efficiency across hybrid cloud environments. • Orchestrated Azure Pipelines on Spark clusters within Databricks, optimizing data processing workflows and enhancing overall system performance. • Deployed solutions in pipelines using Docker and Kubernetes, ensuring scalable and resilient system architecture. • Utilized Azure DevOps for CI/CD pipelines, facilitating streamlined development, testing and deployment processes for increased agility and reliability. • Designed and executed end-to-end pipelines for the ML projects • Collaborated with cross-functional teams to define project requirements, ensuring alignment with business objectives and stakeholder expectations. • Regularly communicated project updates and milestones to key stakeholders

Hedging portfolios under market imperfections using Reinforcement Learning : - State of the art overview - Q-Learning under transaction costs, trading limits, liquidity constraints and costs of funds

- Prototype of a semantic search engine (using Elasticsearch & BERT) - Contribution in deployment of existing PoCs

- Programming a software (.exe using Python + SQL) to automatically manage daily production reports