The sting you never get
is the one that was seen first.

Brazil has seen an uninterrupted, multi-decade rise in scorpion envenomation. Annual cases went from roughly 67,000 in 2014 to more than 200,000 in 2024, and deaths climbed from 76 to over 150 — passing snakebite mortality for the first time. The drivers are ecological: rapid urbanisation without sanitation gives Tityus stigmurus and Tityus serrulatus, both able to reproduce without a mate, ideal habitat; warmer winters speed their feeding and breeding.

In rural Northeast Brazil, antivenom is often more than ninety minutes away. For a child under ten, the outcome of a severe sting depends entirely on how fast it is noticed and treated. OpenScorpion works on the step before that: preventing the sting.

A ruggedised, water-resistant unit mounts at floor level or at an entry point and streams video to an on-device AI model trained on the venomous species of its own region. When a threat is recognised with enough confidence, the unit fires a cascade of alerts within seconds. Everything runs locally — the model, the decision, the alarm — because rural connectivity cannot be trusted with a life.

With the household's consent, each detection can also be logged — species, confidence, GPS, time, conditions — turning every unit into a citizen-science sensor for research partners. First-generation units are alarm-only by design; deterrence hardware exists on the roadmap and is deliberately held back until the detection record is proven.

This is not a concept deck. The training pipeline runs end to end; 42,016 provenance-tracked images across 124 species have been assembled from iNaturalist, GBIF, EOL and Flickr under their licences; and all six Brazilian biomes plus Mexico and North America have trained YOLOv8 detection models, exported to ONNX and smoke-tested, each with its own model card, species list, IBGE- or INEGI-sourced boundary and metrics.

The backend — a European VPS, PostgreSQL, FastAPI, object storage and device-health monitoring — is deployed. A first field unit has been assembled and photographed in a rural kitchen in Paraíba. What remains is the carrier board, the mobile app, certification, and a supervised fifty-unit pilot.

Deployed across a rural region, the units form a living sensor grid: where Tityus stigmurus is expanding its range and how fast; how a La Niña rainfall pattern changes scorpion activity in the Caatinga within seventy-two hours; which house types, micro-habitats and sanitation conditions predict encounters. Questions that have never been answerable at scale, because the only data we have comes from the fraction of encounters that reach a hospital.

The ecological pressure driving scorpions into homes — deforestation, urbanisation, climate disruption — is the same pressure behind a dozen other emerging health threats. A scorpion sensor network is also a probe for how fast shifting ecosystems are pressing on human settlement.

Our working estimate: 20–40% fewer stings in covered households. In a state like Paraíba, several hundred prevented stings a year per ten thousand homes — and potentially dozens of lives.

These projections are directional and should be read conservatively. No system of this kind has been deployed and measured before; generating that real-world evidence is the point of the pilot.

The concept is proven, the architecture is documented, the models are trained and the first unit works. What the project needs now is a fifty-unit supervised pilot in Paraíba, a university co-applicant, and the certification that lets it scale through Brazilian and international grant programmes. We publish the budget because a partner should not have to ask.

In Brazilian reais at ≈ R$ 5.24 per US dollar (March 2026). Imported electronics carry 60–100% in Brazilian tariffs, which the budget reflects. Full line items are in the partner brief.