At Etisalat by e&, I ship production AI/ML models on Databricks across CVM and Technology. For ONT alarms, I built a pipeline that narrowed 213 indicators to about 27 features and captured up to 80% of work orders over four days.
I also improved propensity models by fixing data leakage and duplicated account-month records, and built a customer look-alike recommender using VAE embeddings. A digital-products model reached about 80% accuracy on repeat customers and was adopted for business targeting.
At Manara, the AI reading companion I founded and built, I created a spoiler-safe RAG system that limits retrieval to each reader’s furthest page. I also rebuilt Define on a 13M-row lexicon and developed evaluation and cost-control systems across more than 20 models.
Earlier, at Smarter.ai, I piloted AI solutions that increased web conversions by 31% for Nordgreen, Wayward Collective, and Alyve Wellness. My work has also included computer vision, deployed data science projects, and a Dockerized live Twitter sentiment app.
