Applied Scientist - FinOps

FinOpsly is an AI-native Value-Control platform for cloud (AWS, AZ, GCP), data (Snowflake, Databricks), and AI economics, built to help enterprises move beyond passive cost visibility to active, outcome-driven control. The platform unifies technology spend across cloud infrastructure (AWS, Azure, GCP), data platforms (Snowflake, Databricks), and AI workloads into a single system of action-combining planning, optimization, automation, and financial operations. Role DescriptionMost FinOps tools show you what you spent.

We're building one that tells you what you'll spend, why it's growing, and what to do about it - with models that are evidence-backed and customer-specific, not generic benchmarks. As our Applied Scientist for FinOps, you'll own the science layer of the product: forecasting future Databricks spend, detecting anomalies before they become budget overruns, and powering the optimization recommendation engine that sits at the heart of what we deliver. You'll work embedded in the engineering team, taking models from prototype to production yourself.

What You'll do: Work directly with Databricks System Tables, Unity Catalog metadata, and AWS Cost & Usage Reports to understand the raw signals available. Design and build feature pipelines that transform raw usage and billing data into model-ready inputs - cluster utilization patterns, job frequency distributions, DBU burn rates, instance type mix, and more. Design, train, and evaluate forecasting, anomaly detection, and optimization models using MLflow on Databricks.

Run rigorous experiments - back testing forecasts against historical spend, measuring anomaly precision/recall on labeled incidents, validating savings estimates from optimization recommendations against realized outcomes. Own the path from model to production - packaging models for inference, building scoring pipelines on Databricks, and instrumenting monitoring for model drift, data quality degradation, and prediction accuracy over time. You ship it, you monitor it, you improve it.

Translate model outputs into clear, decision-ready narratives for the product team, engineering leads, and - through the product UI - customers themselves. Define how forecasts, anomaly alerts, and recommendations are explained in the product so they're trusted and acted on, not ignored. Confidence intervals and uncertainty belong in the UI, not just the noteboo What You'll Bring3+ years building and shipping ML models in production - not just notebook experiments handed off to engineers.Strong foundations in time-series forecasting - seasonality, trend decomposition, multi-step ahead prediction, and evaluation methodology.Solid statistical grounding - hypothesis testing, probability distributions, uncertainty quantification, and causal reasoning.Proficiency in Python (scikit-learn, statsmodels, PyTorch or similar) and SQL for data manipulation at scale.Experience with MLflow or equivalent for experiment tracking, model versioning, and deployment.Ability to work with raw, messy billing or operational data and build clean feature pipelines from it.Clear communicator - can explain model behavior, uncertainty, and tradeoffs to engineers and non-technical stakeholders alike.Preferred:Hands-on experience with Databricks - running jobs, understanding cluster configs, using MLflow natively on the platform.Prior work with cloud billing data - AWS CUR, Azure cost exports, or GCP billing datasets.Experience with anomaly detection on operational or financial time-series data.Familiarity with optimization techniques - linear programming, constraint satisfaction, or operations research methods.Knowledge of Bayesian modeling frameworks (PyMC, Stan) for probabilistic forecasting.Background in cloud infrastructure cost modeling or FinOps domain.

Why Join usWe're an early-stage team where your models go directly into the product, the impact is quantifiable in dollars, and there's no committee between your work and the customer.The quality of your forecast is visible in whether customers stay within budget. The precision of your anomaly detection is visible in incidents caught early. You'll have feedback loops most applied scientists never get.

From raw billing data to production inference pipeline. You're not a research function feeding models to an ML platform team - you're the person who takes it all the way. We're deliberate about who we hire - not just for skills, but for how people think, collaborate, and raise each other's work.

You'll be surrounded by people who care about doing things well: clear technical thinking, honest feedback, and a shared intolerance for cutting corners that create future pain. Early teams set the culture for everything that follows - we're building one where senior engineers mentor, not gatekeep, and where good ideas win regardless of seniority.