Why this role exists
Magnasoft is twenty years into building one of the world’s deepest geospatial data assets — and is now turning that
asset into AI-powered software products. Our products turn complex real-world documents and imagery into
structured, usable data using computer vision, OCR, and a human-in-the-loop review loop.
We’re building the small, senior AI team that builds these products. This is one of two core hands-on AI/ML engineer
seats, working directly under our Principal AI Engineer. Important to be explicit up front: this is not a train-a-model-
and-hand-it-off role. You build the models and the product code they live in — the AI backend, the post-processing and
pipeline logic, and the data layer. If not the AI team, no one writes that code. Expect your time to split roughly half
model work, half backend/pipeline work.
What you’ll do
• Build and ship production models — object detection, segmentation, OCR/text extraction, and classification
models behind our products. Not notebooks that die in a repo: models real customers depend on.
• Build the AI backend the models live in. Run the models on incoming data, then write the post-processing
and pipeline logic that turns raw model output into clean, structured product data. All in Python.
• Work in the data layer. Detected and human-corrected results are stored in a document store (MongoDB) —
you design document structures and write the queries and aggregations your pipeline and the retraining loop
depend on.
• Feed the data flywheel — the annotation → correction → retraining loop that makes the models better release
over release.
• Own evaluation for your work — benchmarks, error analysis, and quality metrics tied to real product
outcomes (cost-of-error, reviewer effort saved), not just headline accuracy.
• Deploy and run your models and your pipeline code — Docker, Kubernetes on AWS EKS — and iterate on
what production tells you.
• Work under the Principal AI Engineer’s technical direction, and partner