SRHS: Smart Reptile Husbandry System
IoT and AI proof of concept: a three-layer architecture (Edge, Cloud, Application) combining Azure IoT Hub, computer vision, machine learning and a React/React Native front-end to automate reptile husbandry.
What This Is
The Smart Reptile Husbandry System (SRHS) is a simulated proof of concept for an IoT and AI platform designed to automate, optimise, and safeguard captive reptile environments. It is not a shipped product. The architecture and codebase described here represent a working simulation; no hardware has been deployed.
The project exists to demonstrate end-to-end systems thinking across a domain I know well (reptile keeping), applying the same architecture patterns used in professional digital twin and IoT work at Insight.
Three-Layer Architecture
The system is organised into three layers, each with a distinct technology stack and responsibility:
Edge Layer (Enclosure Hardware). Python on a Raspberry Pi controller, managing sensors (temperature grid, IR camera, gas, humidity) and actuators (lamps, fans, mister). The edge performs local computer vision and data aggregation to minimise cloud transmission cost. Critical life-support functions persist without cloud connectivity.
Cloud Layer (Central Intelligence). Azure IoT Hub for telemetry ingestion, Cosmos DB for data warehousing, Azure ML for the Core Control AI agent, Google Gemini API for the Insight Engine (natural-language alerts, behavioural analysis, system Q&A), and Azure Functions for automation logic. The cloud layer is designed to run the AI control loop, maintain the Digital Twin, and issue commands back to the edge.
Application Layer (User Interface). React web dashboard for advanced reporting and configuration. React Native mobile app as the primary user interface. Both display real-time telemetry, health status, set-point adjustment, and proactive alerts generated by the Insight Engine.
Core Intelligence
Core Control AI. A reinforcement-learning agent designed for Azure ML, using a Digital Twin model of the enclosure. The twin is designed to be continuously calibrated by the sensor grid and used to find energy-efficient control policies for heating, misting, and ventilation. Climate and season simulation models external weather data to adjust internal gradients proactively.
Insight Engine. Designed around the Google Gemini API, this module provides advanced reasoning on top of collected data. It generates natural-language alerts (“Your bearded dragon’s activity level has decreased by 30% over the last 2 days”), analyses IR camera feeds for behavioural patterns, and supports natural-language queries against historical data.
Computer Vision. OpenCV on the edge, processing camera feeds locally. Only derived observations (detected behaviours, anomalies) are transmitted to the cloud, not raw video. A thermal camera (MLX90640) provides non-contact body temperature estimation and basking efficiency metrics.
Safety Architecture
Animal safety is the P0 requirement. The architecture includes redundant temperature sensors with software hard limits and a separate mechanical thermostat as the final fail-safe cutoff. UPS integration sustains essential life support for a minimum of four hours during power loss. Door contact sensors trigger immediate alerts. The AI rejects any sensor reading that contradicts the Digital Twin’s prediction.
Data Model
The system’s data model spans 11 core entities: User, Enclosure, Species, SpeciesCareRule, Animal, Device, MaintenanceLog, TelemetryReading, ActionLog, ObservationLog, Alert, FeedingLog, and HealthEvent. The schema uses JSON fields for semi-structured data where flexibility is needed (device properties, observation context, health notes).
Why This Matters
This project demonstrates the same architectural thinking applied in professional engagements (Lifeblood Azure Digital Twins, DC Two portal dashboard) but with full ownership of every layer. It combines IoT, AI/ML, computer vision, cloud architecture, and front-end development in a single system, and it applies the AI orchestration stance from professional work: choosing the right model for the task (Azure ML for real-time control, Gemini for reasoning and natural language), not defaulting to one tool for everything.
Hard Skills
Python, FastAPI, React, React Native, TypeScript, Azure IoT Hub, Azure ML, Azure Digital Twins, Cosmos DB, Google Gemini API, OpenCV, Raspberry Pi, reinforcement learning, IoT architecture, data modelling
Soft Skills
End-to-end systems architecture, domain expertise applied to technical design, AI tool selection and orchestration, technical documentation
Note: This is a simulated proof of concept. The architecture is described accurately; production deployment is not implied. The current codebase is a minimal simulation representing an early development stage.