Mohamed Afkir

I worked as a Data Analytics Intern within the Data & Analytics team at Yuri & Neil, contributing to data projects for retail and e-commerce clients, including Mes Demoiselles Paris, and IKKS. - Collaborated with client teams to translate business needs into data solutions, from collection and modeling to decision-ready reporting. - Built and maintained ELT pipelines (Airbyte/Mage.ai → BigQuery → dbt) using a Medallion architecture (raw → bronze → silver → gold), orchestrated via Kestra. - Developed Looker Studio dashboards to monitor key business indicators, including revenue and commercial performance. - Optimized GCP/BigQuery queries and data models via partitioning, clustering, and FinOps practices → cut processing costs by 30% and processing times by 60%. - Set up automated Slack/Teams alerts for dbt tests and data quality checks to catch anomalies and pipeline failures before they impacted reporting. - Developed data products for IKKS, including store segmentation models for commercial analysis. - Built a conversational data agent in Google Data Studio, testing its reliability in answering business questions from client data. - Contributed to integrating the agent into Gemini Enterprise, making data-driven analysis accessible to business teams. - Managed task prioritization and client coordination through Jira. Key Results - Saved 6 hours/week in analytics delivery/reporting through automated alerting and dashboard workflows. - Contributed to a 20% increase in tracked revenue through more reliable, actionable dashboards. - Reduced BigQuery processing costs by 30% and processing times by 60% via query and model optimization. - Improved data reliability by monitoring 100+ dbt tests and quality checks through automated alerts. - Deployed a conversational data agent in Gemini Enterprise, enabling business teams to query 25+ BigQuery tables across 3 domains in natural language.

I worked as an AI & Data Science Intern at WASIT, contributing to the development of an autonomous multi-agent system designed to assist Moroccan farmers via WhatsApp. - Designed and developed a multi-agent system for planning, reasoning, and action execution, using Azure Functions for scalable, event-driven orchestration. - Integrated a Retrieval-Augmented Generation (RAG) pipeline to provide reliable, context-aware responses. - Implemented LLMOps monitoring with Langfuse to trace agent interactions, detect response drift, and evaluate generation quality in production. - Designed and managed a PostgreSQL database with pgvector for semantic search and structured tracking of agricultural projects. - Integrated Azure OpenAI, LangChain, Twilio, and Azure Blob Storage across the solution. Key Results - Achieved an average 4-second response time per query for the multi-agent system. - Deployed a scalable, event-driven architecture using 3 Azure Functions. - Enabled contextual, personalized responses through a RAG pipeline drawing on 10+ sources and documents. - Designed and managed a PostgreSQL database supporting semantic search and tracking of 3+ agricultural projects. Tech stack: Python, Azure, Azure Functions, Azure OpenAI, LangChain, PostgreSQL, pgvector, Twilio, Azure Blob Storage, Langfuse, Git, GitHub

I worked as a Data Engineering Intern at Shiftbricks, a startup, where I contributed to building a data pipeline for processing Arabic legal documents and preparing them for AI applications. - Designed a Medallion architecture (bronze → silver → gold) to process and structure Arabic legal documents. - Built a document processing pipeline combining OCR and LLMs to extract and structure metadata and content from PDFs into a standardized JSON schema. - Stored structured data and embeddings in MongoDB to support downstream AI applications, including a legal assistant chatbot. - Automated the pipeline with Apache Airflow, including a human validation step allowing legal experts to review and validate document structures before embedding. - Developed a FARM stack interface (FastAPI, React, MongoDB) for expert validation and integrated the pipeline components for end-to-end processing. - Containerized the entire solution with Docker to ensure reproducible and scalable deployment. Key Results - Automated the end-to-end processing of 80+ Arabic legal documents, from PDF extraction to structured data and embeddings. - Combined OCR and LLMs to extract and structure 15+ fields/metadata per document, refined through human validation. - Prepared structured and embedded content for use in an AI-powered legal assistant. - Delivered a containerized and orchestrated data pipeline using Airflow and Docker.