AI/ML Engineer (Pittsburgh)

Profitmind is a retail analytics SaaS company that turns competitive and customer data into agent-driven insights, helping retailers make faster, sharper merchandising and pricing decisions. Based in Pittsburgh and backed by a recent strategic investment from Accenture, were scaling our agentic AI platform for some of the worlds largest retailers.

The models are the product. Demand forecasts, price and cross-price elasticities, promotion lift, and the optimization that turns them into recommendations are what a merchant actually acts on. This role builds them.

The Role Were hiring an AI/ML Engineer to build and productionize the models behind our insights demand

forecasting, price and cross-price elasticity, promotion and driver modeling, and the constrained optimization that turns those signals into a recommendation a buyer will act on.

The difficulty is rarely model selection. Its that the same model has to hold up across retailers whose data density, category hierarchies, seasonality, and calendars all differ and has to produce output a merchant will believe. Youll own both the statistics and the plausibility: take models from experiment to scheduled, monitored production pipelines, know when a technically valid result is commercially nonsense, and partner closely with product, ML, data engineering, and application teams.

What Youll Do

• Onboard new retailers on our current stack. Bring each customer live on our existing Python forecasting stack rather than rebuilding per customer. Forecast at the grain the business plans at, handling intermittent/low-velocity products, no-history products, and categories where the signal wont support a SKU-level model extending the stack, not re-running it unchanged.
• Model price and cross-price elasticity. Own elasticity with halo, cannibalization, and net impact as separate, inspectable quantities; validate the underlying math against the data, not just the fit.
• Identify what actually drives sales. Own driver identification, going beyond feature selection to corroborate candidate drivers against transaction-level evidence.
• Own pricing across the product lifecycle. Develop logic spanning initial pricing, promotion, markdown, and clearance as one connected problem, incorporating competitive price movement alongside our own demand signal.
• Evolve the optimization engine. Extend our Python optimization to handle real constraints (margin floors, price points, brand standards, inventory/sell-through targets), and ensure the optimizers number matches the number the insight reports.
• Turn methodology into pipelines. Move modeling from notebooks into automated, schedulable pipelines; when an approach fails on a new retailers data, diagnose and generalize it rather than fork it.
• Defend plausibility. Build validation that catches implausible output before customers see it an implausible cross-category lift, a recommendation whose impact contradicts the number beside it, an optimizer that excludes most of a category and be explicit when a fallback to a coarser grain is needed.
• Support explanations, own production, and mentor. Produce the decomposition and driver ranking behind the platforms plain-language recommendations to merchants; own your pipelines in production (scheduling, monitoring, reruns); and bring junior engineers up on the methodology.

What Were Looking For

• Production ML, not notebooks 5+ years. Hands-on building and shipping ML systems in Python (scikit-learn, PyTorch, statsmodels, or similar), owning models running in production.
• AI-assisted engineering. Daily use of agentic coding tools and AI-driven automation, with the seniority to judge when a tools output is wrong.
• Forecasting depth. Time-series and demand forecasting at scale, including intermittent/sparse demand methods (Crostons or comparable), hierarchical reconciliation, seasonality detection, and cold-start.
• Elasticity and causal inference. Price/cross-price elasticity, promotion lift measurement, and the discipline to separate correlation from causation control selection, quasi-experimental design, holdouts.
• Optimization under constraints. Applying mathematical optimization (linear programming, constraint satisfaction, or similar) to real business rules.
• Genuine engineering ability. Strong Python and expert SQL (parent-child hierarchies, store clusters); can take a model from experiment to a scheduled, monitored pipeline and debug it at customer scale; comfortable with distributed processing (Spark or similar), containers, and orchestration.
• Retail mechanics. Understands markdowns, promotions, inventory turn, margin, and sell-through well enough to recognize a retail-implausible result even when error metrics look fine. Prior retail, pricing, revenue management, or supply chain experience strongly preferred.

Nice to Have

• Bayesian or hierarchical modeling for sparse retail data
• Retail-specific modeling: markdown/clearance optimization, assortment, open-to-buy, or product matching
• LLM-generated narrative or agentic systems built on top of quantitative model output
• Databricks/Delta Lake, Spark at scale, and workflow orchestration (Airflow, Argo, or comparable)
• MLOps practice: model versioning, monitoring, drift detection, and backtesting
• Advanced degree in ML, statistics, econometrics, or operations research; experience as an early ML hire

Location Hybrid - Pittsburgh, PA. Works across the ML, data engineering, application, and product teams.

Why This Role Most ML roles hand you a metric to improve. This one hands you the decisions a retailer makes about price, promotion, and inventory, and asks you to make the models behind them both accurate and believable. Its early enough that the methodology and the standard for a trustworthy result are still yours to set.

Profitmind is an equal opportunity employer. We welcome applicants of all backgrounds.