Machine Learning Engineer — Predictive Retail Models (Amazon SageMaker)

WHAT WE'RE LOOKING FOR We need someone with hands-on experience building predictive models on Amazon SageMaker using proprietary (non-public) datasets. You've done this in the real world — not just tutorials or Kaggle competitions. You've wrangled messy company data, trained models, and validated predictions against what actually happened. We're also looking for your input on how to scope the MVP for this phase. We have the data, we have the business context, and we have a clear end goal — but we want your expertise in shaping what the minimum viable proof-of-concept looks like. We'll discuss this in more detail with shortlisted candidates. THE PROJECT We are a consulting company focused on helping small businesses adopt AI into their workflows. Our client is also a consulting company: they advise retailers how to adjust their merchandising targets to maximize their sales and profits. We are building the tools that the consultants will use to create client reports and recommendations. But the big gaping hole is that they're basing their advice on prediction models that are run on spreadsheets. There's a lot of real world experience embedded in those workbooks, so it'd be silly to dismiss them out of hand. But we think that (just like in a lot of other domains) a ML model that's trained on all of their data can do even better. Your job in this phase is to answer one question: Can we create purpose-built models that make better forecasts than the current spreadsheets or "off the shelf" models like Claude or OpenAI? Specifically: - Help define the MVP scope — what's the smallest meaningful proof-of-concept that tells us if this approach is viable? - Build predictive models using Amazon SageMaker trained on historical retail data - Backtest model predictions against actual historical outcomes - Compare multiple modeling approaches to determine what works best for this problem space - Define and apply validation criteria — what level of accuracy makes a model trustworthy? - Communicate results clearly to non-technical stakeholders REQUIRED EXPERIENCE - Proven experience building and deploying ML models on Amazon SageMaker (not just experimentation — production or near-production work) - Experience working with private/proprietary datasets (not exclusively public or academic data) - Time-series forecasting or demand prediction experience - Model validation and backtesting methodology - Comfort with messy, imperfect real-world data NICE TO HAVE - Retail or supply chain domain experience - Experience with multi-variable "what-if" scenario modeling - Familiarity with inventory management concepts (turns, weeks of supply, markdown optimization) BROADER CONTEXT This is Phase 1 of a larger initiative. If models prove valid, subsequent phases will expose predictions to consultants as an interactive tool — allowing them to experiment with different parameter combinations and see probable outcomes before making recommendations to their retail clients. This role builds the predictive foundation that everything else depends on. IMPORTANT: Applications that do not include a compelling real-world example in response to the screening question below will not be reviewed. Freelancers only; no agencies.