Description

We are seeking a data scientist with strong statistical intuition and practical forecasting experience to build simulation-aligned predictive models in a planning environment. This role requires designing probabilistic models to support time-sensitive decisions involving sparse, volatile, and clustered event data. You’ll work closely with simulation engineers and planners to ensure that model outputs can drive downstream rollouts with quantified confidence, not just point estimates. The ideal candidate understands that the goal is not just “good forecasts,” but reliable forecasts that guide high-stakes operational decisions under uncertainty. You will shape the probabilistic foundation used to guide large-scale simulation of multi-day plans, making trade-offs between risk, opportunity, and feasibility.

Job duties:

Design and implement probabilistic forecasting models over discrete time bins (e.g., daily, 8-hourly) using historical count data

Model likelihoods and path success probabilities to support downstream rollout and routing simulations

Build cluster-aware models using hierarchical priors or partial pooling to handle sparse, imbalanced geographic data

Engineer features based on seasonality, time-of-week, type, and operational filters

Construct and calibrate prediction intervals (e.g., 90-95% credible ranges) to support confidence-aware path scoring

Quantify the risk of failure when chaining uncertain sequences of events, using Monte Carlo methods or weighted path aggregation

Collaborate with rollout engineers to ensure model outputs integrate smoothly with scoring pipelines and employee logic

Validate models contextually — not just with RMSE/MAE — but in terms of their value in simulation or planning objectives

Track and mitigate forecast degradation in slow, skewed, or volatile clusters

Skills:

• Strong applied statistics background:
  • Poisson, Negative Binomial, Zero-Inflated models
  • Overdispersion modeling
  • Hierarchical or partially pooled models
• Demonstrated experience forecasting counts or rare events over time and space
• Proven ability to model uncertainty, not just point forecasts — includes quantile regression or probabilistic calibration
• Proficient with LightGBM / XGBoost for both classification and regression, especially under class imbalance
• Ability to integrate forecasts into Monte Carlo-style simulations, scoring downstream path viability and value
• Experience crafting statistically meaningful features from calendar, recency, and volume patterns
• Strong understanding of how models are used in planning, not just how they perform in isolation
• Fluent in designing confidence-aware models, especially for scenarios where success depends on sequential availability
• Familiarity with spatiotemporal data structures, and with evaluating risk vs. reward under incomplete information

Preferred (Not Required):

• Bayesian modeling frameworks (e.g., PyMC3, Stan)
• Familiarity with quantile forecasting or distributional prediction
• Experience building models for operational planning, routing, or scheduling systems
• Knowledge of score calibration techniques (e.g., Platt scaling, isotonic regression)