ML Engineer

What you'll do

• Build feature pipelines

on EMR Serverless Spark and Event streaming that turn lake data into versioned, point-in-time-correct behavioral features per user and per identity.

• Own feature management:

feature definitions, registry, backfills, freshness and drift monitoring, and making sure training and inference read the same features.

• Train and evaluate models

for insider-risk scoring and anomaly detection: data prep, labeling with the analyst team, experiment tracking, and evaluation on precision, recall and alert volume.

• Deploy and serve models

on the Intuit ML Platform: batch inference for the daily composite score, real-time inference through

MXS

(Model Execution Service) and

MSaaS

, and LLM-based enrichment through

LXS

(LLM Execution Service).

• Keep scores explainable:

per-feature contributions and evidence links that an analyst can follow back to source events.

• Run the ML lifecycle:

retraining schedules, model versioning, shadow and A/B rollout against the rule-based score, monitoring, runbooks and on-call.

• Work with the data engineering and analytics teams

to move the application from weighted rules to ML scoring across FY27, including non-human identity (NHI) scoring.

Must-have skills

• Production data pipelines:

built and operated batch and streaming pipelines with Spark/PySpark and Kafka.

• ML training:

trained, evaluated and shipped models to production (gradient boosting, anomaly detection, sequence or graph models), with sound evaluation and attention to label quality.

• Feature management:

designed and run feature stores or feature pipelines, including point-in-time joins, feature versioning, training/serving consistency and backfills.

• Model serving and MLOps:

deployed models behind batch and real-time inference, with CI/CD, model registry, monitoring and rollback.

• Python

as primary language (production services, PySpark, pytest), plus strong

SQL

and data modeling on Iceberg/Hive tables with Athena or Databricks.

Data analysis:

profiling messy telemetry, checking signal quality, and validating score distributions against expectations