The Get Lost Lab

Everything we build is first refined inside a fully operating beachfront resort property.

TECHNICAL SHOWCASE: THE ENGINEERING BLUEPRINT

Network Overview & Hardware Infrastructure

The entire estate relies on an enterprise-grade network backbone designed for high availability and zero downtime in a harsh coastal environment. Network routing and security are anchored by a UniFi Dream Machine Pro (UDM Pro) paired with a US-24 PoE 250W managed switch, delivering robust power over Ethernet to all critical devices. Due to the expansive, multi-structure layout of the resort, ten strategically positioned Ubiquiti access points—including advanced U7 Pro units in high-traffic zones like the main house patio and balcony, and U6 models in outlying areas—blanket the property in a seamless wireless mesh. To overcome distance limitations and eliminate signal degradation across the estate, MikroTik managed switches handle high-speed fiber-optic distribution directly to the Beachfront and CSA zones. This infrastructure is continuously monitored in real time by the central automation controller, which uses smart polling filters to ignore temporary system restarts while instantly dispatching telemetry alerts via encrypted channels if an access point goes offline.

System Architecture Diagram Description

At the core of the property's intelligence is a completely self-hosted, cloud-independent architecture powered by Home Assistant running on dedicated local hardware. This centralized "brain" acts as a universal translator across disparate hardware ecosystems, localizing data that would traditionally be sent to third-party cloud servers. The architecture is split into three core layers: data ingestion, local edge processing, and hardware execution. The ingestion layer continuously pulls telemetry from the UniFi network, the professional weather station, energy inverters, and security cameras. The local processing layer features dedicated hardware accelerators that handle facial recognition, machine learning vision models, and automation logic entirely on-premises. Finally, the execution layer translates these decisions into physical actions, controlling everything from the property's ten independent air conditioning systems to whole-home audio distribution.

Unified Dashboard Design

The user interface is built on highly optimized, contextual Lovelace dashboards designed to turn massive amounts of technical telemetry into clean, actionable insights at a glance. The Weather Dashboard replaces generic forecast icons with live, high-precision gauges, an integrated real-time Windy.com radar map embed, and a dynamic wind compass. The Energy Dashboard provides a clear, transparent view of the property's 14.3 kW solar matrix, displaying real-time photovoltaic production, battery state of charge, and net-metering grid imports/exports, while instantly converting kilowatt-hours into a tangible financial metric in Philippine Pesos. For daily operations, the Main House and Resort dashboards offer granular control over climate, lighting, and presence tracking, complemented by physical, in-wall touch panels running customized blueprints for instant, smartphone-free adjustments.

Weather Forecasting & Smart Shell Logic

The local weather forecasting engine operates entirely independent of external internet-reliant APIs, utilizing a physical, professional-grade GW2000A weather station mounted on-site. Every 15 minutes, a custom dew point depression algorithm runs locally, calculating the micro-climatic gap between temperature and dew point alongside rapid variations in barometric pressure and humidity trends. The system uses this mathematical analysis to classify local rain risk into five precise tiers ranging from "none" to "imminent." Because the physical structures of the resort are rotated exactly 27.5 degrees clockwise from true north, the system maps this orientation against real-time wind vector data from the anemometer. If the algorithm determines that rain is likely, it audits the physical open/closed states of the estate's eight Zigbee-monitored bifold doors, instantly matching the incoming wind direction to the open openings and sending a targeted notification naming the exact rooms requiring physical closure.

Secure Local AI Systems & Video Analytics

The property’s spatial awareness is driven by a private, local artificial intelligence stack that processes data with absolute privacy. Built around a high-performance local network video recorder, the security system continuously analyzes eleven high-definition camera streams without uploading a single pixel to the cloud. When motion occurs, an on-premises facial recognition engine detects known family members, guests, or strangers, while a vehicle sub-labeling model identifies specific cars, such as the family SUV, by make, model, and license plate. Simultaneously, local vision language models hosted via Ollama evaluate camera snapshots to generate concise text descriptions of events. A custom filtration pipeline automatically suppresses false positives caused by wind, shadows, or insects; if the vision model detects an empty frame, the notification is discarded, ensuring the family only receives high-value alerts.

Automated Resort Operations & Automation Flows

Resort operations and guest experiences are governed by complex, multi-conditional automation flows that adapt dynamically to property occupancy. Using virtual boolean helpers and Google Calendar integrations, the system continuously tracks booking statuses across the main house, guest suites, and individual glamping tent sites. When a campsite is occupied, the property alters its security perimeter profiles, updates light paths, and pathways voice alerts to respect guest privacy while maintaining security. If the weather station flags an incoming tropical storm or a dangerous wind advisory, the automation flow branches automatically: it cross-references which tents are occupied and sends targeted safety notifications specifically to the relevant operational staff and guest zones. These interactions are governed by a master security mode selector that overrides standard automation behaviors depending on whether the estate is in Home, Guest, Night, Away, or Alarm status.

Hyper-Local AI Weather & Event Summaries

The ultimate layer of refinement on the property is the deployment of a locally-hosted language model that humanizes raw sensor data into natural language summaries. Every fifteen minutes, a script packages the current meteorological readings—including temperature, humidity, UV index, solar irradiance, and barometric trends—and passes them to an on-site text reasoning model. The model interprets these numbers and outputs a single-sentence description of the current conditions, which is displayed on the main dashboard card and updated throughout the day. This data-to-text synthesis also powers the daily Morning Briefing; at 10:00 AM, the automation engine converts the local AI's plain-English weather synthesis, the current solar harvest efficiency, and system operational metrics into a high-quality text-to-speech broadcast, delivering a hands-free, screen-free status update across the estate's media speakers.