I need an experienced data-science engineer, already familiar with telecommunications data, to build a production-ready customer-churn prediction model in Python. Because the work involves direct access to our internal systems, you must be based in Jakarta and available for occasional on-site sessions with the analytics team. What you will work with Our data lives in a relational database and combines three rich sources: customer demographics, detailed usage patterns, and historical service-log events. I will grant read-only credentials so you can extract, transform, and merge the tables as you see fit. Scope of work – Explore and clean the raw tables, handling missing values and outliers that are common in telco records. – Engineer features that capture behavioural and tenure signals decisive for churn. – Train, tune, and compare several Python models (e.g., XGBoost, LightGBM, or any proven churn-focused algorithm). – Evaluate against a held-out set; I’m targeting an AUC of 0.80 or higher, with clear lift over our current rule-based approach. – Package the final pipeline—pre-processing plus model—so it can be scheduled nightly. A simple Flask or FastAPI endpoint for real-time scoring is a plus. Deliverables 1. Jupyter notebook (or .py script) covering data prep, modelling steps, and explanation of key decisions. 2. Exported model artefacts and requirements.txt for reproducibility. 3. Short read-me describing deployment instructions and expected inputs/outputs. 4. One hand-over session in Jakarta to walk through results and answer questions. If you have proven telco projects in your portfolio and can start quickly, let’s discuss the timeline and any access you’ll need.