Amine Slimane

Technical leadership and architecture for the AI solution processing client files, from design to industrialization, to transform and accelerate risk analysis. 🔹 𝐌𝐨𝐝𝐮𝐥𝐚𝐫 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞: Designed a multi-stage processing pipeline using a fleet of Qwen-VL family models of varying sizes. Each task is assigned to the lightest and fastest model meeting the requirements, drastically optimizing overall speed and computational costs without any compromise on accuracy. 🔹 𝐃𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐏𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧: Orchestrated the analysis chain, from automatic document classification and page indexing, to high-fidelity data extraction from scanned documents. 🔹 𝐏𝐫𝐨𝐚𝐜𝐭𝐢𝐯𝐞 𝐂𝐨𝐧𝐭𝐫𝐨𝐥 𝐚𝐧𝐝 𝐀𝐥𝐞𝐫𝐭 𝐒𝐲𝐬𝐭𝐞𝐦: Developed cross-validation modules that automatically raise alerts for case managers when inconsistencies are detected between documents (e.g., name discrepancies, address mismatches) or file completeness risks. 🔹 𝐃𝐞𝐜𝐢𝐬𝐢𝐨𝐧 𝐒𝐮𝐩𝐩𝐨𝐫𝐭 & 𝐓𝐫𝐚𝐜𝐞𝐚𝐛𝐢𝐥𝐢𝐭𝐲: At the end of the pipeline, generation of an auditable summary document that revolutionized the initial verification process. This report includes the final risk score, with automatic alerts for high-risk levels and the dynamic hyperlinks pointing directly to the source of each piece of information in the original documents, drastically reducing manual verification time. 🔹 𝐑𝐨𝐛𝐮𝐬𝐭 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧: Deployed this solution on on-premise infrastructure using Llama.cpp (performance and structured generation), combined with quantization and GPU offloading techniques to ensure smooth execution on existing hardware.

𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐋𝐞𝐚𝐝 for the end-to-end design and industrialization of an AI solution to evaluate Chanel products' online presence across e-commerce sites: 𝟖𝟎% 𝐦𝐚𝐢𝐧𝐭𝐞𝐧𝐚𝐧𝐜𝐞 𝐞𝐟𝐟𝐨𝐫𝐭 𝐫𝐞𝐝𝐮𝐜𝐭𝐢𝐨𝐧 and 𝟏𝟎𝟎% 𝐝𝐚𝐭𝐚 𝐫𝐞𝐥𝐢𝐚𝐛𝐢𝐥𝐢𝐭𝐲. 🔹 𝐂𝐨𝐧𝐟𝐢𝐠𝐮𝐫𝐚𝐭𝐢𝐨𝐧-𝐃𝐫𝐢𝐯𝐞𝐧 & 𝐌𝐨𝐝𝐮𝐥𝐚𝐫 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞: Developed an architecture decoupling business logic (Python) from configuration (YAML), enabling non-technical teams to add sites, pages, and user journeys (clicks, hovers) without coding. 🔹 𝐂𝐨𝐧𝐭𝐞𝐧𝐭-𝐁𝐚𝐬𝐞𝐝 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐭 𝐒𝐞𝐥𝐞𝐜𝐭𝐨𝐫𝐬: Implemented robust element trackers relying on semantic content rather than HTML structure, ensuring resilience against website changes and drastically minimizing maintenance requirements. 🔹 𝐀𝐈-𝐏𝐨𝐰𝐞𝐫𝐞𝐝 𝐕𝐢𝐬𝐢𝐛𝐢𝐥𝐢𝐭𝐲 𝐈𝐝𝐞𝐧𝐭𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐒𝐜𝐨𝐫𝐢𝐧𝐠 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞 : Deployed across 20+ international sites with 300+ scrapes/week (1,500+ pages and 10,000+ elements analyzed weekly) via multiple iterations on critical sites. Brand identification using Google Cloud Vision API (Logo Detection + Product Search for image similarity) and custom scoring algorithm based on communication type (group vs brand). Dual quality validation and screenshot archiving reduced incidents from 20+/week to less than 4/week.

Contributed to the design and deployment of the group's strategic Smart Data Hub on Google Cloud Platform, serving 70 airports across 14 countries (318M passengers/year). 🔹 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐃𝐚𝐭𝐚 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞: Production deployment of a centralized platform unifying several business domains (air traffic, operations, finance, retail, HR, communication) with daily processing of several TB of data. 🔹 𝐀𝐈-𝐀𝐮𝐠𝐦𝐞𝐧𝐭𝐞𝐝 𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞: Development of an automated data quality system (Great Expectations) coupled with an intelligent alerting mechanism using LLMs to generate diagnostics and correction recommendations. 🔹 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐈𝐦𝐩𝐚𝐜𝐭: Core infrastructure for the Executive Committee's decision-making dashboards and for the industrialization of Generative AI use cases by the Data Science team.