Data Scientist (ML, Forecasting) (Mumbai)

Role & Responsibilities Technical: Python fluency. Daily-driver level. pandas, numpy, scipy, matplotlib. Comfortable innotebooks and in modular code. Time series forecasting. Hands-on with at least: ETS / Holt-Winters, ARIMA, Croston (or similar intermittent-demand methods). You know what temporal cross-validation is and why standard k-fold breaks on time series. You can explain why MAPE breaks on zero-inflated data and what to use instead (WAPE, MASE). Statistical intuition. You know when to be suspicious of a model that fits too well. You can spot data leakage. You instinctively check for stationarity, seasonality, and structural breaks before fitting anything. Inventory or supply-chain math literacy. Even if not your day job you understand or can pick up fast: safety stock, reorder point, EOQ, service level / fill rate, (s,S) policies, lead-time variability. You don't need to derive them; you need to read a formula and know which assumption is doing the work. Monte Carlo / simulation comfort. You can vectorize a simple inventory simulation in numpy without reaching for a framework. You understand bootstrap, sampling distributions, and how to read a simulation result. EDA discipline. You start every dataset with the same questions: row count, null rate, dtype, distribution, time coverage, key uniqueness. You produce a one-page "what's in this data" before you fit anything. How you work: Hypothesis-driven. Comfortable being given "I suspect X, go check" rather than a spec. Comfortable coming back with "actually, the data shows Y . Iterative and visual. Charts before tables, tables before paragraphs. You'd rather show than tell. Honest about uncertainty. "I don't know yet; let me get back to you in 2 days" is a excellent answer. Over-confidence on shaky numbers is an undesirable trait in this role. Self-directing on the day-to-day, while welcoming senior input on direction. you can't be waiting for them. Role & Responsibilities Technical: Python fluency. Daily-driver level. pandas, numpy, scipy, matplotlib. Comfortable innotebooks and in modular code. Time series forecasting. Hands-on with at least: ETS / Holt-Winters, ARIMA, Croston (or similar intermittent-demand methods). You know what temporal cross-validation is and why standard k-fold breaks on time series. You can explain why MAPE breaks on zero-inflated data and what to use instead (WAPE, MASE). Statistical intuition. You know when to be suspicious of a model that fits too well. You can spot data leakage. You instinctively check for stationarity, seasonality, and structural breaks before fitting anything. Inventory or supply-chain math literacy. Even if not your day job you understand or can pick up fast: safety stock, reorder point, EOQ, service level / fill rate, (s,S) policies, lead-time variability. You don't need to derive them; you need to read a formula and know which assumption is doing the work. Monte Carlo / simulation comfort. You can vectorize a simple inventory simulation in numpy without reaching for a framework. You understand bootstrap, sampling distributions, and how to read a simulation result. EDA discipline. You start every dataset with the same questions: row count, null rate, dtype, distribution, time coverage, key uniqueness. You produce a one-page "what's in this data" before you fit anything. How you work: Hypothesis-driven. Comfortable being given "I suspect X, go check" rather than a spec. Comfortable coming back with "actually, the data shows Y . Iterative and visual. Charts before tables, tables before paragraphs. You'd rather show than tell. Honest about uncertainty. "I don't know yet; let me get back to you in 2 days" is a excellent answer. Over-confidence on shaky numbers is an undesirable trait in this role. Self-directing on the day-to-day, while welcoming senior input on direction. you can't be waiting for them.