ePhi — custom satellite-imagery algorithm under $90 per metropolitan area

The challenge

Process satellite imagery covering large industrialised areas across multiple French cities. Analyse every square metre to identify buildings as potential commercial-development targets. Manual review wasn't feasible at this volume; building a full in-house AI ML pipeline was too costly and time-intensive for the timeline they had.

The solution

Google Maps was the right source — consistent image scale, predictable resolution, and per-metro costs under $90. The Static API isn't designed for ML, so we wrote a custom algorithm that captures, refines, and enriches imagery in real time, then feeds it into a structured pipeline ready for downstream models.

Deployment ran on AWS to handle the long-running, rate-limited scraping process. Images are stored in S3, ready for SageMaker or any other ML service to pick up.

The outcome

A working capture-and-process pipeline delivered inside the two-month window the client had set. Imagery flows in with consistent parameters and predictable costs, and emerges as structured data that downstream analysis can act on.

The same pattern is reusable for any geographic-imagery task that requires consistent scale and bulk processing without bespoke ML infrastructure.