Lead AI/ML Engineer – Remote (India)

• Location
• Remote – India

About Supy

Supy is transforming how restaurant groups manage inventory, procurement, and decisions. We’re building an agentic AI layer on Databricks that not only powers insights but also takes safe, automated actions.

Why this role matters

As our Lead AI/ML Engineer, you will be at the heart of this transformation—designing production-grade ML platforms, building forecasting systems, and shaping how agentic AI drives real business outcomes at scale. This is not just about research; it’s about deploying AI that directly impacts thousands of restaurants worldwide.

What you’ll do

• Architect ML platforms: Build and maintain feature stores, forecasting pipelines, and reusable ML assets on Databricks.
• Deliver business-critical forecasts: Design probabilistic, hierarchical demand forecasts with exogenous drivers (holidays, events, promotions).
• Build agentic AI integrations: Wire models into agents (RAG + tool-calling) that can power replenishment, anomaly detection, and production planning.
• Productionize models at scale: Deploy models with Databricks Model Serving, APIs, and schema-validated tools.
• Own observability: Track forecast accuracy, model drift, and system health with dashboards and automated checks.
• Mentor & lead: Publish playbooks, guide engineers, and co-own ML reliability with the Data Engineering team.

Who you are

• Graduate from IIT or NIT.
• Hands-on experience in Databricks (production use): Unity Catalog, Delta Lake, MLflow, Model Serving, Feature Tables, Vector Search.
• Strong background in forecasting & ML: ARIMA/ETS/Prophet, GBMs, TFT, DeepAR, hierarchical reconciliation, quantile forecasts.
• Advanced skills in PySpark, Spark SQL, and Python.
• Strong mindset for production readiness: APIs, schemas, governance, lineage, and safety.
• Bonus: Experience with LangChain/LangGraph, optimization problems, or hospitality/retail AI solutions.

In your first 90 days, you’ll

• Launch baseline forecasts with accuracy dashboards.
• Deploy governed feature tables reused by BI and agents.
• Ship one production model powering an end-to-end agent.