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, TensorFlow, or PyTorch.
- 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 BigQuery (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, NeuralProphet, TabPFN, Chronos); probabilistic and Bayesian modeling; AutoML tooling such as PyCaret 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 BigQuery.
- 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 (LangGraph, 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.
