ML Engineer

ABOUT DOGWOODLINE:

We make a web-based inventory and purchasing system used by independent restaurants, breweries and small food manufacturers. Founded in 2010, the business writes the billing, metering and customer portal software that small water and electric utilities run on.

THE ROLE:

Dogwoodline is looking for an ML Engineer to join a small, steady team. This hire will help build out our ML platform: feature store, training pipelines, model registry and deployment. The Fraud team trains and serves models that run on every customer request. We need someone who cares about latency as much as accuracy. Remote position. Most of the team is in Berkeley, but nobody is required to come in. Normal working hours, Monday to Friday. Please make sure you have a quiet place to work and a solid internet connection.

WHAT YOU'LL WORK ON:

• Monitor models for drift, data quality issues and latency regressions.
• Cut inference cost with quantization, batching and caching.
• Productionize data science prototypes without rewriting everything.
• Write clean, tested Python and review teammates' code.
• Document model cards, data sources and known failure modes.

WHAT WE'RE LOOKING FOR:

• 2+ years of software engineering experience, with at least two spent on ML systems in production.
• Familiarity with Docker, Kubernetes and a cloud ML platform (SageMaker, Vertex AI, Azure ML)
• Understanding of core ML concepts: overfitting, evaluation metrics, bias in training data.
• Knowledge of LLM APIs, prompt design and evaluation methods.
• Hands-on experience with PyTorch, TensorFlow or JAX.
• Strong Python and solid engineering habits (testing, packaging, code review)

PLUSES:

• Computer vision background (OpenCV, detection, segmentation)
• Distributed training with DeepSpeed or FSDP.
• CUDA or Triton kernel experience.

PAY AND BENEFITS:

• Health, dental and vision coverage starting on day one.
• 401(k) with a 100% match up to 5%.
• Three weeks of PTO to start.

Start: Within 30 days.

Demo job created on 11jobs for testing and demonstrations. It is not an actual opening at Dogwoodline.