Machine Learning Engineer

Role Overview

We are looking for a hands-on ML Engineer to join our engineering team. You will own the end-to-end design, implementation, and operation of the customer-facing predictions platform: from raw behavioral data through feature engineering, model training, serving, monitoring, and retraining. This is Localytics's first major ML-powered product capability, and this person will work directly with the Head of Product to bring it to life.

You will be expected to stand up the ML infrastructure yourself: training pipelines, feature store, model registry, and serving layer, with support from Cloud and Platform Engineers on the underlying cloud infrastructure.

The core pipeline processes ingested behavioral, event, and marketing data through cleaning, transformation, feature engineering, model inference, and output generation, delivered on a recurring offline schedule and via real-time inference. The predictions and other models built here will play a major part in the next-best-action decisioning system.

This role partners closely with the CTO, Head of Engineering, and Head of Product on architecture, infrastructure, and product decisions.

About Localytics

Localytics is an AI-enablement firm for enterprise marketing organizations. We build, integrate, and operate marketing AI for modernizing teams.

We have been working inside CRM marketing for fifteen years. We have a long history of successfully supporting enterprise brands through our mobile analytics and engagement platform. Today, our mission is to help modernizing brands integrate the latest technologies across their marketing workflow, help them operate it, dramatically impacting outcomes.

Localytics is a lean, fast-growing team at an exciting juncture. We are building a work culture that values integrity, ownership, and impact. No matter your role, you'll have real responsibility, influence, and the opportunity to build things that matter.

What You'll Do

Own the Pipeline End to End

• Take full responsibility from raw feature consumption through model deployment and production monitoring
• Treat model degradation as a first-class incident, monitoring drift proactively, investigating root cause, and shipping improvements on a regular cadence

Deliver the Predictions API as a Product

• Work directly with the Head of Product on the Predictions API as the primary near-term outcome, understanding that prediction quality is a product quality signal, not just an ML metric
• Partner daily with the CTO, Head of Engineering, and Head of Product on architectural and product decisions, and work closely with Cloud Platform Engineers on feature pipelines, serving infrastructure, and orchestration

Work AI-Forward

• Use AI coding assistants to accelerate model development, pipeline authoring, and documentation
• Design and deploy AI agents to reduce manual toil across the ML lifecycle, including feature validation, retraining triggers, and experiment summarization

Capabilities You Bring

ML Foundations

• Supervised learning: deep knowledge of learning algorithms for customer behavior modeling, including logistic regression, gradient boosting (XGBoost, LightGBM), neural networks, and ensemble methods. Strong intuition for algorithm selection given data shape and prediction objective
• Probabilistic and statistical modeling: understanding of uncertainty quantification, probability distributions, and statistical inference as they apply to prediction problems. Familiarity with time-series forecasting and causal reasoning is a plus
• Prediction problems: hands-on experience with prediction problems directly applicable to mobile engagement, such as churn prediction, channel propensity, and message affinity scoring

End-to-End ML Ownership

• ML lifecycle: experience owning the complete ML lifecycle, including data generation, feature engineering, training pipeline design, model evaluation, registry, deployment, monitoring, drift detection, and retraining
• Predictions API: able to design and implement a versioned, low-latency Predictions API consumed by external customers, not just internal tooling

ML Infrastructure: Self-Sufficient Setup

• SageMaker: able to provision, configure, and operate AWS SageMaker end to end, including Training Jobs, Pipelines, Model Registry, Feature Store, and real-time Endpoints, largely independently, without relying on a dedicated MLOps team. Cloud and Platform Engineers will assist with underlying AWS infrastructure but will not own the ML stack
• Orchestration: experience setting up and operating enterprise data orchestrators such as Apache Airflow, Prefect, or Kubeflow Pipelines for scheduling and managing ML workflows
• Spark: able to write, debug, and optimize Spark pipelines for data cleaning, transformation, and feature computation at scale
• Data lakes: conversant with data lakes (S3, Parquet, Delta Lake, Apache Iceberg) and how raw event data flows from ingestion through to model-ready features

Feature Engineering & Stores

• Feature stores: experience working with feature stores (SageMaker Feature Store or equivalent), consuming offline features for training, online features for low-latency inference, and collaborating on feature definitions and historical backfills
• Behavioral features: strong feature engineering instincts for behavioral time-series data, including recency, frequency, trend, and engagement pattern features derived from event streams

Programming

• Proficient with ML libraries: scikit-learn, MLlib, PyTorch or TensorFlow, and Hugging Face where applicable
• Comfortable with SQL
• Familiarity with Golang or Java is a plus
• Experience with MLflow or SageMaker Experiments for run tracking, metric comparison, and reproducibility

AI and Data Governance

• Bias detection: awareness of model bias detection and mitigation, understanding how prediction models for audience selection and engagement can encode or amplify bias, and applying evaluation techniques to surface and address it
• Explainability: familiarity with model explainability techniques to make prediction outputs interpretable for product teams and customers
• Regulatory compliance: working knowledge of data governance principles and regulatory considerations (GDPR, CCPA) as they apply to customer behavioral data used in ML pipelines

Nice to Have

• Experience building stateful, cyclic AI workflows using LangGraph or similar frameworks (LangChain, CrewAI)
• Experience with multi-task or joint prediction models, predicting churn and channel propensity simultaneously from shared representations
• Experience with A/B testing and online evaluation frameworks for model rollout in production
• Exposure to mobile engagement, adtech, or marketing automation platforms and the behavioral data they generate
• Familiarity with a Lakehouse architecture as a data source upstream of the feature pipeline
• Containerization experience (Docker, Kubernetes/EKS) for packaging and deploying model serving endpoints
• Familiarity with Model Context Protocol (MCP) and how prediction endpoints can be exposed as MCP-compatible tools so AI agents can consume them directly, consistent with an API-first platform architecture
• Experience writing AI agents to automate ML pipeline operations

Outcomes You Will Deliver

Ship the Predictions API

• Predictions API: join the team and contribute immediately to pipeline design, delivering a production-grade, versioned Predictions API within the first three months, the team's first major ML customer outcome
• End-to-end ML pipeline: design and build the full pipeline, from ingestion and cleaning of behavioral and event data through transformation, feature engineering, model training and inference, and structured output generation

Own the ML Infrastructure

• ML infrastructure ownership: provision and operate SageMaker training pipelines, Spark ETL jobs, orchestration via Airflow or Prefect, model registry, feature store, and serving endpoints from greenfield, keeping them well-documented, reproducible, and maintainable
• Agentic ML ops: reduce manual operations overhead by deploying agents with automated telemetry that handle drift alerting, retraining triggers, experiment reporting, and efficiency

Keep Models Accurate and Reliable

• Model quality: establish accuracy baselines for each prediction type and improve them continuously through systematic experimentation, feature iteration, and scheduled retraining
• Drift detection and reliability: monitor all production models for data and concept drift, catching degradation early and maintaining prediction quality SLOs as customer behavioral distributions shift over time

Deliver Product and Platform Value

• Short-term product impact: enable campaign audience selection powered by churn probability, channel affinity, and fatigue signals, giving Localytics customers a meaningful new capability
• Long-term platform value: contribute to the broader next-best-action decisioning system as a core member of its build. The Predictions API is the starting point, and this engineer will continue to expand the intelligence layer as additional decisioning components are built around it