Mouad Hamri

I designed and deployed an enterprise-grade agentic AI assistant for the compliance and audit department. The system enables the teams to query regulatory documents, policies, and audit reports through natural language, powered by a self-hosted LLM stack deployed on Azure AKS. The architecture combines LangChain-based agentic workflows, Qdrant vector search for RAG retrieval, and Langfuse for observability and evaluation. The assistant autonomously retrieves relevant regulatory content and synthesizes answers, significantly reducing manual document review time. The solution was built with a focus on data privacy, using fully self-hosted open-source LLMs with no data leaving the organization's infrastructure.

I architected and implemented a centralized Microsoft Fabric Lakehouse for consolidating data from multiple source systems, including Salesforce, Odoo, and operational databases. The solution applies a full medallion architecture (bronze, silver, and gold layers) for data ingestion, quality control, and transformation using PySpark and Delta Lake. The gold layer powers enterprise-grade Power BI dashboards, enabling real-time reporting and operational decision-making across departments. The platform significantly reduced data silos and enabled cross-team data sharing with a unified, governed data model.

I designed and deployed an end-to-end agentic document processing pipeline for automated invoice data extraction. The solution combines self-hosted OCR models with open-source LLMs served via vLLM, orchestrated through LangChain agents to extract structured information from unstructured invoice documents. Extracted data is validated using Pydantic schemas and outputs clean, structured JSON ready for direct integration with accounting and ERP systems. The pipeline eliminated manual invoice processing, significantly reducing processing time and human error across the finance department.