Senior AI/ML Engineer

UMATR; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

UMATR in San Francisco is seeking its first dedicated ML Engineer to own models end-to-end, from data to production. You will work with large operational data and 500k+ SKUs, building systems that directly impact how the business operates.

This is not a research role; you will build, deploy and operate production ML systems and take ownership when reality changes. You will own forecasting models, data pipelines, MLOps tooling, and potentially contribute frontend work with TypeScript/React to

Qualifications

• 5–7 years of experience building production ML systems.
• Experience building time-series forecasting models serving production traffic.
• Hands-on experience with AWS SageMaker.
• Experience fine-tuning LLMs using LoRA or PEFT on real datasets.
• Experience building systems backed by ontologies or knowledge graphs.
• Strong experience engineering large-scale data pipelines with Spark.
• Experience owning production models through deployment, monitoring, drift detection and retraining.
• Strong architecture skills and the ability to explain technical decisions in detail.

Responsibilities

• Build production forecasting models across messy, intermittent and seasonal demand, including cold-start SKUs, promotions, perishability and long-tail demand.
• Build datasets and fine-tune models using LoRA / PEFT, with rigorous evaluations determining what actually ships to production.
• Build the representation layer that allows AI systems to reason across inconsistent products, vendors, pack sizes and units of measure.
• Own the infrastructure around those models, including deployment, versioning, monitoring, drift detection and automated retraining.
• Build large-scale ML and data workloads using Python + Spark.
• Work with AWS SageMaker, S3, Glue + Step Functions.
• Build production inference and evaluation infrastructure.
• Use MLflow, Kubeflow or equivalent MLOps tooling.
• Contribute outside the model layer when needed, including enough TypeScript/React to work across the wider product.

Skills

Time-series forecasting

Production ML systems

Python

Spark

AWS SageMaker

LoRA/PEFT

MLOps tooling

Knowledge graphs

TypeScript

React

Data pipelines

Model deployment

Tools

MLflow

Kubeflow

SageMaker

Job description

We are working with a fast-growing AI startup building the operating brain for the supply chain. They’ve grown 10x in the last year with a small engineering team and are now building out the model layer underneath their production AI systems.

They’re looking for their first dedicated ML Engineer to own models end-to-end, from raw data through to production. You’ll work with years of real-world operational data across 500k+ SKUs, building systems that directly impact how the business operates.

This is not a research role, and it’s not an LLM-wrapper role. They’re looking for someone who can build, deploy and operate production ML systems - and take ownership when reality changes.

What you'll own

• Build production forecasting models across messy, intermittent and seasonal demand, including cold-start SKUs, promotions, perishability and long-tail demand
• Build datasets and fine-tune models using LoRA / PEFT, with rigorous evaluations determining what actually ships to production
• Build the representation layer that allows AI systems to reason across inconsistent products, vendors, pack sizes and units of measure
• Own the infrastructure around those models, including deployment, versioning, monitoring, drift detection and automated retraining
• Build large-scale ML and data workloads using Python + Spark
• Work with AWS SageMaker, S3, Glue + Step Functions
• Build production inference and evaluation infrastructure
• Use MLflow, Kubeflow or equivalent MLOps tooling
• Contribute outside the model layer when needed, including enough TypeScript/React to work across the wider product

There are no handoffs. You’ll build the model, put it into production, monitor it and fix it when reality changes.

What we're looking for

• 5-7 years of experience building production ML systems
• Experience building and maintaining time-series forecasting models serving production traffic
• Hands-on experience with AWS SageMaker
• Experience fine-tuning LLMs using LoRA or PEFT on real datasets
• Experience building systems backed by ontologies or knowledge graphs
• Strong experience engineering large-scale data pipelines with Spark
• Experience owning production models through deployment, monitoring, drift detection and retraining
• Strong architecture skills, with the ability to explain and defend technical decisions in detail
• Comfortable working across the full ML lifecycle rather than owning just one part of the process

They’re looking for someone who can talk about what happened after the model shipped - when it degraded, how you detected it, what it got wrong and what you changed.