Higharc Machine Learning Engineer builds and ships ML models that understand homes, owning problems end to end from research to product deployment. Responsible for training, evaluating, serving models, and maintaining latency and cost budgets.
Responsibilities
Our team builds AI that understands homes, and our generative floor plan engine has moved from proof of concept to a core workflow in Higharc's product. You'll own the applied side of that work. The research team hands you a defined problem and a metric. You train and adapt the models, build the pipelines around them, and ship a component that meets the product's latency and cost budget.
You'll work inside an architecture that already exists. The engine, the schema, and the evaluation frameworks are in place, and your job is to make them faster, reproducible, and easy for others to build on.
Build the component layer around our layout synthesis engine: a documented API contract, a service the product can call, and an evaluation gate that runs on every change.
Turn model retraining into a one-command job, with benchmarks built in and results the whole team can read.
Own the latency and cost budgets for learned capabilities.
Extend core research into new capabilities, such as multi-story plans and real-time editing.
Partner with infrastructure engineers on MLOps and deployment, and mentor researchers and interns on engineering practices.
Share your work in writing through weekly status updates, clear pull requests, and design notes before big changes.
Qualification
Fluent in PythonMaster's degree in computer scienceA major plus if you also bringAn architecture or building background
Required
You're happiest when a model you worked on is running in the product. You like taking a well-defined problem with a metric attached and closing it, and you improve an existing codebase without feeling the need to rewrite it. You think about what a 45-second wait means for a user before you think about what it means for a benchmark. You're honest about evaluation: you can tell when a result is an artifact of the test set, and you'd rather report a regression in writing than hide it behind a demo.
Most of your collaborators are remote and some are part-time, so writing things down comes naturally to you. You can go from a research conversation to a product conversation to an infrastructure conversation in the same week and be understood in all three. You also know enough about buildings to see that a bedroom door opening onto a kitchen is wrong before any metric tells you.
Five or more years of professional software development, at least two of them training and shipping machine learning models to real users rather than in notebooks.
Has carried at least one model end to end, from data preparation through training, evaluation, serving and the integration that put it in front of users.
Fluent in Python; PyTorch preferred, TensorFlow or Keras acceptable with a commitment to move.
Has worked with geometric or structured data such as floor plans, CAD or BIM, graphs, or 2D and 3D geometry.
Has built or maintained an evaluation harness or benchmark and used it to gate changes. Has packaged a model behind an API (FastAPI or similar) and been accountable for its latency and cost.
Comfortable working remotely with a team on US Eastern hours.
Master's degree in computer science, machine learning or a related field, or equivalent shipped work.
A major plus if you also bring:
An architecture or building background, whether a degree, practice, or AEC software work.
Experience with constraint solvers such as OR-Tools CP-SAT or with optimization more broadly. Generative models for layouts or floor plans specifically. Plugin or tooling work in Rhino, Grasshopper or Revit. Hugging Face ecosystem, experiment tracking, and cloud model serving on AWS or Modal. Publications or open source contributions. A PhD is not required.
