The Platform
Not a camera trap with an app. A production-grade AI infrastructure: edge vision processing, multi-agent threat assessment, secure mesh networking, and a 30-agent fleet that never clocks off. Built to prove itself on one property in South Africa — and then license to every game reserve, national park, and wildlife trust that needs it.
Why Existing Tools Fail
Conservation's standard response to poaching is rangers, patrol schedules, and community engagement. All of these matter. None of them watch a 200-hectare property simultaneously at 3am on a Tuesday.
Traditional camera traps generate enormous volumes of images — the majority of which are wind-triggered vegetation, passing livestock, or non-target wildlife. A ranger reviewing camera trap footage in the morning cannot act on a threat that occurred at midnight. The latency between detection and response is the gap that poachers operate in.
Existing commercial wildlife monitoring systems require cloud connectivity, ongoing subscription costs, and hardware that does not survive South African bush conditions. They are designed for European and North American operators, not for a smallholding in Limpopo with intermittent power and no guaranteed cellular signal.
The Alpha-Panga platform is built for field realities: edge processing only, hardware that runs on solar, alerts that reach rangers via WhatsApp or radio bridge, and a cost structure that conservation organisations can actually sustain.
How It Works
Five layers. Each one purpose-built. Together they form a detection pipeline that closes the latency gap between threat and response.
Layer 1
IP cameras and thermal imagers feed via local LAN into Frigate NVR. RTSP streams, 24/7 recording with motion detection zones.
Layer 2
Coral TPU accelerator classifies each motion event in real time: pangolin / human / vehicle / other wildlife. No cloud. No latency.
Layer 3
Multi-model LLM fleet evaluates classified detections: time of day, location, movement pattern, proximity to burrow sites.
Layer 4
High-confidence poaching threats trigger immediate ranger notifications — WhatsApp, radio bridge, or direct call based on operator preference.
Layer 5
Every detection logged with timestamp, confidence score, and footage clip. Forensic record for prosecution. Population data as a byproduct.
The Stack
Each component of the stack is open-source or commercially available hardware. The integration is Alpha-Panga's IP — and it is what makes the platform deployable in environments where commercial solutions have never operated.
Vision Infrastructure
Open-source network video recorder purpose-built for AI detection. Processes RTSP streams from any IP camera. Hardware-accelerated detection zones, configurable sensitivity per zone, motion event recording. Runs on-site — no footage leaves the property.
Local processing · Any IP camera · Configurable detection zones
Edge AI Inference
Google's Coral Edge TPU processes TensorFlow Lite models at the hardware level — 4 TOPS of inference performance from a USB dongle or M.2 module. Classification runs at the camera site, not in a data centre. Models are customised for local wildlife conditions: SA bush species, night-vision characteristics, typical human intrusion patterns.
4 TOPS inference · USB/M.2 form factor · Offline capable
Threat Intelligence
The Alpha-Panga agent fleet (30+ specialised AI models) provides contextual threat assessment that raw classification cannot. An unaccompanied human at a fence line at 2am is different from a ranger doing a perimeter check. Context — time, location history, movement vector, proximity to known pangolin burrow sites — is what separates a real alert from a false positive.
RTX 6000 Ada · 48GB VRAM · qwen3-vl, llama4:scout, deepseek-r1
Secure Networking
Remote properties without fixed IP addresses or reliable cellular coverage connect via Tailscale — a WireGuard-based mesh VPN. Every camera node, every edge device, and the central processing cluster form a secure private network with no exposed ports. A property in Limpopo connects to the central fleet as securely as a device on a corporate LAN.
WireGuard encryption · No fixed IP required · Zero exposed ports
The Differentiator
Alpha-Panga is not a conservation organisation that bought some cameras. It is a technology company that built a commercially deployed AI infrastructure — and then pointed part of that infrastructure at wildlife crime.
The same fleet that ships SABS Suite v2 (a Prokon-replacement structural engineering platform for South African engineers) is the fleet that processes pangolin camera feeds. The same GPU that runs multi-model structural calculations runs vision inference on camera trap images. This is not a side project — it is an integral use of existing production infrastructure.
30+ specialised AI agents, all running on South African infrastructure. Each has a defined role. Together they form a system no single model can replicate.
| Agent Role | Conservation Application | Commercial Application |
|---|---|---|
| Vision classification | Camera trap image analysis, poacher detection | Structural element detection, drawing analysis |
| Natural language | Alert drafting, incident reporting, researcher queries | Client communication, report generation |
| Data analysis | Population trend modelling, movement corridor mapping | Engineering load analysis, compliance checking |
| Scheduling | Ranger patrol optimisation, sensor maintenance scheduling | Project timelines, deadline management |
| Web / SEO | Conservation reporting, donor communications | Commercial lead generation, content marketing |
Revenue & Sustainability
Phase 1 proves the platform. Phase 2 and beyond license it. The model that funds pangolin conservation long-term is not donor dependency — it is technology licensing revenue from the conservation organisations, private game reserves, and government wildlife agencies that need exactly what Alpha-Panga has built.
High-value wildlife properties protecting rhino, elephant, and pangolin face sophisticated poaching networks. The Alpha-Panga stack deploys in 2–4 weeks. Per-property annual licensing covers ongoing model updates and support.
South African National Parks manages 21 national parks. Scaling edge AI detection across remote park sections is a solvable infrastructure problem — and a procurement opportunity for proven technology.
Phase 3 and 4 expand to Central Africa and Asia. International conservation NGOs operating in DRC, Cameroon, Indonesia, and the Philippines will be the licensing partners that bring the platform to those species.
Academic research institutions and IUCN working groups benefit from aggregated detection data (anonymised, location-generalised). Data partnerships generate goodwill, citations, and grant co-funding opportunities.
Licensing enquiries: partner@alphapanga.com
Deployment Roadmap
Phase 1 — Active
Single property in Gauteng/Limpopo. Full stack deployment. Temminck's ground pangolin. Platform validation against live threat environment.
Phase 2
3+ partner properties. Tailscale mesh connecting distributed sites. Platform handles 200+ hectares simultaneously from single control point.
Phase 3
Adapt platform for arboreal species (white-bellied pangolin). DRC, Cameroon NGO partnerships. Models retrained for rainforest camera conditions.
Phase 4 & 5
All 4 Asian species covered. Platform licensed globally. Revenue model self-sustaining. Technology potentially open-sourced to global conservation community.
Get Involved
Whether you manage a game reserve, run a conservation fund, or want to partner at the corporate level — talk to us.
Corporate & Licensing Enquiries Read the Full Mission