Founding Geospatial Machine Learning Engineer

Homecastr; New York (NY); Hybrid; USD 295,000 - 325,000; Full time

Job summary

Homecastr is seeking a founding Geospatial Machine Learning Engineer to transform our spatial forecasting model into a production-ready system. You will own end-to-end modeling from design to deployment, with a focus on rigor, calibration, and explainability across regional indicators.

The role requires leadership in probabilistic forecasting, geospatial data handling, and reproducible training pipelines, aiming to deliver practical, tested solutions for public-sector and industrial applications.

Qualifications

• Six or more years in applied ML, scientific computing, geospatial modeling, forecasting, or a related field, with staff-level ownership or equivalent evidence.
• Strong Python and modern ML framework experience, including production model development.
• Skill with probabilistic or time-series evaluation, uncertainty, calibration, or comparable statistical rigor.
• Experience with geospatial data, coordinate systems, spatial joins, geographic hierarchies, and large spatial datasets.
• Experience building reproducible training and evaluation systems rather than notebook-only analysis.
• Ability to move between research questions, data engineering, model implementation, and production constraints.
• Clear scientific writing and the judgment to state limitations precisely.

Responsibilities

• Design, train, evaluate, and deploy forecasting models across parcels, buildings, neighborhoods, infrastructure, utilities, and regional indicators.
• Build leakage-resistant historical datasets with explicit vintages, geographic crosswalks, target definitions, and reproducible feature construction.
• Establish benchmarks across places, horizons, baselines, and public planning models.
• Measure point accuracy, probabilistic scores, calibration, coverage, tails, geographic transfer, and failure modes.
• Build uncertainty estimates and explanations that are technically defensible and useful to practitioners.
• Develop and test intervention-conditioned or scenario models without overstating causal identification.
• Own experiment tracking, model lineage, data quality checks, training reproducibility, and model cards.
• Partner with the product engineer to deploy models through stable services with monitoring, cost controls, and rollback capability.
• Work with public-sector practitioners and independent reviewers to turn domain criticism into better datasets, tests, and model behavior.
• Communicate methods and limitations clearly in technical documents, customer materials, and diligence artifacts.

Skills

Python

ML frameworks

Production models

Probabilistic evaluation

Geospatial data

Reproducible training

Tools

GeoPandas

PostGIS

GDAL

xarray

rasterio

Job description

Founding Geospatial Machine Learning Engineer

Turn our physical development model into rigorously validated, production-ready spatial forecasting and scenario capability.

The roadmap spans probabilistic forecasting, geographic transfer, historical-vintage controls, cross-jurisdiction benchmarks, calibration, explainability, multi-target modeling, and intervention-conditioned scenarios. That scientific and engineering responsibility should not remain indefinitely concentrated in the founder.

What you will own

• Design, train, evaluate, and deploy forecasting models across parcels, buildings, neighborhoods, infrastructure, utilities, and regional indicators.
• Build leakage-resistant historical datasets with explicit vintages, geographic crosswalks, target definitions, and reproducible feature construction.
• Establish benchmarks across places, horizons, baselines, and public planning models.
• Measure point accuracy, probabilistic scores, calibration, coverage, tails, geographic transfer, and failure modes.
• Build uncertainty estimates and explanations that are technically defensible and useful to practitioners.
• Develop and test intervention-conditioned or scenario models without overstating causal identification.
• Own experiment tracking, model lineage, data quality checks, training reproducibility, and model cards.
• Partner with the product engineer to deploy models through stable services with monitoring, cost controls, and rollback capability.
• Work with public-sector practitioners and independent reviewers to turn domain criticism into better datasets, tests, and model behavior.
• Communicate methods and limitations clearly in technical documents, customer materials, and diligence artifacts.

FIRST 90 DAYS

Establish the foundation

• Reproduce the current principal benchmark from source data through published metrics.
• Audit target definitions, vintages, leakage controls, geographic joins, and baseline comparability.
• Define the model evaluation contract for one-year and multi-year horizons.
• Produce a prioritized research and engineering plan tied to the first paid evaluation.
• Ship one material improvement to model performance, calibration, geographic coverage, or evaluation reliability.

6 TO 12 MONTHS

• A reproducible multi-jurisdiction benchmark supports customer and investor diligence.
• Forecast and uncertainty metrics are monitored by geography, horizon, cohort, and target.
• New data sources can be added through documented, tested spatial and temporal contracts.
• Models move from experiment to production through a controlled and observable release process.
• The first paid evaluations have independent technical review and defensible acceptance evidence.

What we are looking for

• Six or more years in applied machine learning, scientific computing, geospatial modeling, forecasting, or a related field, with staff-level ownership or equivalent evidence.
• Strong Python and modern ML framework experience, including production model development.
• Skill with probabilistic or time-series evaluation, uncertainty, calibration, or comparable statistical rigor.
• Experience with geospatial data, coordinate systems, spatial joins, geographic hierarchies, and large spatial datasets.
• Experience building reproducible training and evaluation systems rather than notebook-only analysis.
• Ability to move between research questions, data engineering, model implementation, and production constraints.
• Clear scientific writing and the judgment to state limitations precisely.

Helpful, not required

• Public records, land use, transportation, infrastructure, utilities, climate, demography, or economic forecasting.
• PyTorch, distributed training, spatial databases, GeoPandas, xarray, rasterio, GDAL, PostGIS, or equivalent systems.
• Work with planners, government analysts, regulated industries, or independent technical reviewers.

Role boundary

This is not a pure data-engineering position, remote-sensing-only position, or academic research appointment. You must improve model capability, evaluation credibility, and production delivery together.

Compensation and working terms

$310,000 target base salary, 5% target variable compensation, and a 0.75% target equity grant under the current financing plan. Final terms will be confirmed if the role opens.

This role opens after sufficient financing, an upsized close, or initial paid commercial evidence. Location terms will be confirmed when it opens.

Homecastr considers candidates based on relevant evidence, judgment, and ability to do the work. We welcome strong candidates whose path does not match every conventional credential.