Data Scientist

Data Scientist : Forecasting & Optimization (Rust)Company Description Merchmix is a merchandising and inventory operating system with intelligence at its core, designed for retailers, brands, wholesalers, and inventory-led businesses that need sharper control over stock, planning, trading, suppliers, stores, and execution. The platform unifies merchandise financial planning, budget and OTB management, range and assortment planning, allocation and replenishment, visual merchandising, and AI-driven decision intelligence into one connected source of truth. By replacing fragmented spreadsheets and disconnected systems, Merchmix enables teams to move from reactive firefighting to proactive planning, improving inventory precision, protecting margins, and responding faster to demand shifts. The solution creates a single source of truth across teams, regions, stores, suppliers, and channels, integrating with leading enterprise systems including NetSuite, SAP S/4HANA, Microsoft Dynamics 365, Shopify, and BigQuery. ERP remains the system of record, supported by approval-first publish-back and a fully auditable trail.Role Description This is a full-time, on-site Data Scientist role based in Bengaluru. The Data Scientist will work closely with product, engineering, and customer-facing teams to design and implement data-driven solutions that enhance forecasting, inventory optimization, and merchandising decisions across the Merchmix platform. Day-to-day responsibilities include collecting and preparing data from multiple enterprise systems, building statistical and machine learning models, conducting exploratory data analysis, and creating clear visualizations and reports that support business decisions. The role involves experimenting with new algorithms, validating model performance, and translating complex findings into actionable recommendations for non-technical stakeholders. The Data Scientist will also contribute to improving the platform’s AI decision intelligence capabilities and help establish best practices for data quality, governance, and documentation.What you'll own Forecasting Demand at SKU-store-week granularity, most series sparse and intermittent. New product cold-start with no sales history. Hierarchical reconciliation so store forecasts sum coherently to region and national plans. Size and colour curves. Cannibalisation and halo across a range. Seasonality that follows the retail calendar, not the Gregorian one.Optimization and simulation A forecast nobody acts on is a dashboard. You'll build the models that decide what moves: allocation and replenishment, safety stock policy, assortment and range planning, markdown timing and terminal stock risk. And the simulation layer behind our scenario tools — when a merchandiser asks what happens if they shift 10% of stock online, something has to answer credibly and fast.Models that ship as systems You take a problem from definition through prototype to production Rust.Applied AI and agentic systems Our platform runs autonomous workflows, not just analysis. You'll work on the reasoning layer — retrieval over retail context, tool use, natural-language explanation of forecasts — and on the evaluation harnesses that tell us whether any of it works. We care more about someone who can measure an agent's failure modes than someone who can wire one up.Causal measurement Promotional lift, price elasticity, and the counterfactual every merchandiser asks: what would have sold if we hadn't discounted?Evaluation infrastructure, data quality and governance Rolling-origin backtesting, honest baselines, error metrics that reflect commercial cost rather than statistical convenience. We'd rather ship a model we can defend than one that demos well — and since our outputs write back into customers' ERPs, "defend" is literal. You'll help set the standards for data quality checks, model documentation and auditability that let a retailer trust an automated purchase order.Growing into the depth You'll learn retail merchandising from people who ran it at scale, and systems engineering from people who built for it. Two years here should make you materially harder to replace than two years spent tuning models in a notebook — and as the India team grows, the person who was here first sets how it works.Qualifications Required3–5 years building and deploying ML in production. What you shipped matters more than your title.Depth in at least two of: time series forecasting, mathematical optimization (LP/MIP, constraint programming), simulation, or causal inference. All four is rare and we won't pretend otherwise.Genuine software engineering ability, not scripting. You write tested, reviewed, maintainable code. You're comfortable with static types, and you understand ownership, allocation and concurrency well enough that Rust's model will be a new syntax rather than a new concept.Evidence you've learned something hard and unfamiliar, fast — a new language, a new domain, a new paradigm — and shipped with it. Tell us that story in your application; it's the thing we'll dig into.Strong SQL. Window functions and query plans, not just joins.Experience with messy operational data from enterprise source systems — ERP exports with backdated corrections, missing weeks, hierarchies that change mid-year, and a schema nobody documented. If you've reconciled data across SAP, NetSuite or Dynamics, say so.Clear communication with non-technical stakeholders. You'll explain a forecast to a Head of Buying placing $100million+ USD of stock.LLM systems in production, and specifically evaluation — eval sets, regression measurement, cost and latency management, guardrails.