> Markdown version of [/jobs/ext/4638402-founding-machine-learning-data-engineer](https://www.wearedevelopers.com/jobs/ext/4638402-founding-machine-learning-data-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Founding Machine Learning / Data Engineer

- **Company:** Primitive, Inc.
- **Location:** New York, NY, United States
- **Experience:** Expert
- **Contract:** Permanent contract
- **Skills:** Airflow, Amazon Web Services, Amazon S3, Information Engineering, Data Warehousing, Identity and Access Management, Python (Programming Language), PostgreSQL, Machine Learning, Open Source Technology, Operational Databases, Raw Data, Management of Software Versions, Feature Store, Data Ingestion, Pytorch, Build Management, HuggingFace, Machine Learning Operations, Vertica, Drift Detection
- **Published:** November 8, 2026
- **Apply:** https://www.juju.com/job/28_25303_f0381adc09e3a5d1e1dfab70dd7eac93

## About the Role

* 5+ years in ML / data engineering, with at least one production ML system you took from raw data to served predictions
 * Strong Python; comfortable with modern ML tooling (PyTorch, Hugging Face, Ray, or equivalents - we're not religious)
 * Real opinions about data versioning, feature stores, and experiment tracking - you've used DVC, LakeFS, MLflow, or Weights & Biases and know what each is good and bad at
 * Production data pipeline experience with a real data warehouse - schema design, contracts, ownership
 * AWS chops: S3, EKS-hosted training (or SageMaker), IAM that doesn't terrify the security team
 * Built or operated a human-in-the-loop labeling workflow
 * Comfortable setting architectural direction for ML/data at a small company, balancing vision with pragmatism, * Background in hardware, EE, or anything physical - helps a lot when the data is from real machines
 * ClickHouse, dbt, Airflow / Dagster / Prefect at production scale
 * Contributed to open-source ML or data tooling

## Description

Running CI on real hardware produces a kind of data most ML teams never get to touch: power traces, sensor streams, build artifacts, console logs, test results, all tied to specific physical machines doing specific physical work. There's serious signal in there - predicting flaky boards, classifying failure modes, scoring job risk, and eventually closed-loop optimization of how the fleet itself is used.

We're hiring a founding ML / data engineer to build the pipeline that turns that data into models, and the models into product features. You'll own this end-to-end - not "hand off a notebook and hope." Data ingestion, labeling, training infrastructure, evaluation, deployment, monitoring. You'll set the technical direction for ML at Primitive and shape what good looks like - from the first models in production to ML as a core part of the product.

This is the first dedicated ML hire. It's a build-from-zero role on top of a rich, real-world dataset.

What you'll do

 * Design and build the data platform end-to-end: extend instrumentation where signals are missing today, then ingest from Postgres, SeaweedFS / S3, and streaming telemetry into a clean, versioned analytical layer
 * Build the labeling workflow that lets us (and eventually customers) label hardware events without it becoming a permanent side project
 * Design and operate a reproducible training stack on AWS - distributed where it needs to be, with experiment tracking, dataset versioning, and a real eval harness
 * Ship inference for product features: low-latency serving where it matters, batch scoring where it fits
 * Operate models in production: drift monitoring, regression gates, the dashboards that tell us when a model is silently rotting
 * Partner with the full-stack and hardware teams to integrate predictions cleanly into the product surface
 * Set the bar for ML rigor at Primitive: eval-first development, reproducibility, honest reporting of model quality
 * Mentor new engineers on data and ML patterns as the team grows; raise the bar on data contracts, eval design, and reviewability