The trade-off
Higher recall can mean more false positives and more review work. We accept that trade-off to reduce missed smoke detections; false alarms still have an operational cost.
Pantera
The wildfire management platform I lead at 1.5°C, with our engineering, ML, and environmental science team — from risk analysis and smoke detection to response and impact assessment.
- 190 cameras in the network — 150 HaaS (hardware we run for landowners) and 40 SaaS (our software layered on infrastructure that's already there).
- 40M+ hectares under watch, including the Pantanal and indigenous territories.
Pantera — smoke detection
Actual system output
Smoke 97.2%
Same system, live
Raw footage from the camera network. The system patrols preset views, flags a smoke candidate, and zooms in for another check. Detections that cross the confidence threshold go to our 24/7 team for review.
Pantera — live camera feed
Same detection pipeline
Why we prioritize recall
Missing real smoke can delay the response to a wildfire. In this application, that false negative carries a greater cost than a false positive. We prioritize recall in the smoke detection model: finding more of the real smoke occurrences.
From detection to alert
A model detection is a candidate for validation. The camera checks and human review described above are part of the path to a confirmed alert. Model precision and the quality of alerts reaching the customer must be evaluated separately.
What it does
Detection
The camera network and computer vision models flag signs of smoke for validation and early response.
Risk
Weather, fuel, and terrain layered together to flag where a fire is likely to start or spread next.
Response
Supporting the coordination of a field response once a detection is confirmed.
Impact
Measuring what changed afterward — area burned, land recovered — the case for doing this at all.
How it's built
A team of 8 — engineering, geospatial, and data science — running on Agile and OKRs. I've owned the architecture end to end as co-founder and CTO since 2020.
Python
FastAPI
AWS
Docker
PostGIS
PyTorch
Angular
umgrauemeio.com →
Get in touch
If you lead an early-stage climate or govtech team and want a technical second opinion, or you're building something in wildfire or geospatial monitoring, write to me.