At Infor, I serve as technical lead on CoreDeskAI, an internal multi-tenant conversational-AI platform that routes requests across LLM providers and uses RAG-based knowledge retrieval. I improved retrieval calibration across French, Arabic, and Tunisian Derja, and developed testing for prompt injection and tenant data isolation.
I also built Infor’s Revenue Ops Agent, coordinating specialized agents to generate diagnostic reports and reduce manual analysis time by 80%. My Mini-RAG architecture combines asynchronous processing with hybrid search, improving retrieval precision by 25%.
For Infor’s hotel-demand forecasting systems, I developed data pipelines using Spark, SQL, Kafka, and MySQL, with clustering and MAPE-based validation. These systems contributed to a 12% RevPAR uplift and an 8% occupancy improvement.
My earlier work includes borrower and interest-rate models at Acredius and computer-vision and semi-supervised learning systems at Atladev. I also created AgentEval, an open-source LLM evaluation engine, and have hands-on experience fine-tuning models with LoRA/QLoRA and optimizing inference.
