Job description
Job Description WHAT YOU’LL CHAMPION: Duties and Responsibilities Own a major modeling area - forecasting or pricing - and drive multi-quarter initiatives across models, pipelines and people. Work across the following areas: Validation standards: set the backtesting, calibration and uncertainty checks others follow, measured in business terms Experimentation: design and debug live experiments - set them up correctly, size them properly, catch broken readouts before they mislead anyone Demand measurement: measure how customer demand responds to business decisions, and evaluate new policies safely before rollout Production ML: design systems so that what was trained is exactly what runs in production Mentor junior data scientists through review, pairing and honest feedback.
Reframe business asks into well-posed analytical problems - and push back on ill-posed ones. Communicate results and risks honestly to technical and non-technical stakeholders. Able to work under pressure and change, and balance among speed, reliability, interpretability.
WHO YOU ARE
Requirements and Qualifications BS/MS/PhD in a Business, IT, Mathematics, Science or Engineering discipline 4-8 yrs relevant experience beyond first degree, owning production ML that drove real commercial decisions Expert Python and SQL on large time-ordered datasets. Deep validation craft: you build evaluation and monitoring infrastructure, not just use it.
You know when not to add complexity, and can defend simplicity to stakeholders. Comfortable communicating results through stakeholder-facing dashboards (Looker Studio or similar) as well as written analysis. Good working knowledge of productivity tools such as G Suite, Git, Jira, Confluence. Experience from any data-rich industry is welcome — e-commerce, fintech, logistics, telco, ride-hailing or beyond.
Airline or pricing background is not required. Experience in one or more of the following specialized areas: Machine Learning Solid understanding of machine learning algorithms — XGBoost, LightGBM, neural networks, decision trees — with a clear grasp of why you tuned what you tuned. Strong Python and hands-on experience with ML frameworks such as scikit-learn, Tensor.
Flow, or Py. Torch. Demonstrable understanding of forecasting and regression pitfalls — lag feature leakage, target leakage in cross-validation, high-cardinality categorical handling, and the trade-offs between MAE, MAPE, and RMSE — you catch these unprompted and choose objectives and metrics that match the business decision.
Ability to interpret models — SHAP, partial dependence, residual diagnostics — and explain results to non-technical stakeholders without dumbing them down. Statistical modeling beyond tree ensembles — GLMs (Poisson, Tweedie, logistic), quantile regression, hierarchical / mixed-effects models and state-space time-series models — with judgment on when they beat gradient boosting.
Hands-on Google Cloud Platform experience, particularly Big. Query (window functions, partitioning, cost-aware SQL) and Vertex AI (training jobs, model registry, endpoints, pipelines). Nice-to-have: deep learning for tabular and time-series problems (TFT, N-BEATS, Neural. Prophet, TabPFN, Chronos); probabilistic and Bayesian modeling; AutoML tooling such as Py.
Caret for rapid baselining. m Pricing & Demand Science Propensity / take-up modeling with well-calibrated probabilities — you know an uncalibrated probability must never feed a price. Price-elasticity and demand-response estimation from experimental and observational data. Price optimization under business constraints and guardrails, and offline simulation of a pricing policy before it touches customers.
Encoding domain structure into models — e.g. monotonic price-demand constraints in gradient boosting. Nice-to-have: multi-armed / contextual bandits or reinforcement-learning approaches to pricing; personalization and segmentation. Forecasting & Optimization Forecasting large families of related time series — per-product, per-market — with hierarchical structure, seasonality and event effects, and sensible cold-start handling for new products or markets.
Probabilistic forecasting: quantile and distributional forecasts, prediction intervals, and evaluating them honestly (coverage, pinball loss) rather than only point accuracy. Forecasting demand that accumulates toward a deadline (booking- or order-curve style problems), and handling censored demand — when sell-outs truncate what you can observe.
Mathematical optimization on top of forecasts — allocating scarce inventory or capacity via linear / integer programming and marginal-value reasoning. Turning distributions into decisions: expected marginal value of the next unit of inventory (probability of sell-out × expected revenue), quantile-based allocation of remaining capacity, always with business-rule guardrails on top.
Nice-to-have: clustering and market-segmentation methods; simulation-based evaluation of decision policies. Experimentation & Causal Inference You have measured real effects with modern causal methods — double ML, uplift modeling, difference-in-differences, synthetic control, instrumental variables — and know their assumptions and failure modes.
You recognize correlation-vs-causation traps in commercial data unprompted, and propose credible ways to measure the true effect. You can design, size, monitor and analyze A/B tests end to end, and detect a broken experiment from its readout. Algorithm Engineering Experience productionizing models end-to-end — from SQL feature pipelines to deployed serving endpoints — on GCP using Vertex AI and Big.
Query. Orchestrating batch ML on Airflow / Cloud Composer — containerized jobs, sensible schedules and dependencies, alerting that reaches a named owner — and Spark for heavy post-processing when the data demands it. Integrating model outputs with third-party vendor and reservation systems, with data-freshness and sanity checks that fail loudly rather than ship stale decisions.
You have maintained, debugged and retired production pipelines, not just handed models over the wall. Monitoring discipline — drift detection, data quality checks, model performance tracking in production. Nice-to-have: decision-making under uncertainty (inventory/stocking problems); operations research; LLM-based or agentic tooling (Lang.
Graph, MCP servers, eval harnesses). WHERE YOU’LL GO: Dispatcher to captain, ramp agent to data analyst, brand executive to CEO - these are some Dare To Dream stories of our Allstars. Based on your performance and contribution in this role, you’ll grow into becoming a Lead Data Scientist. In this role, you’ll own the team’s modeling portfolio, set methodology and standards, and coach a team of data scientists.