PartSnap - our camera-based equipment identification feature - launched as a tool to tell you what you are looking at. Point your camera at a pump, and it tells you the make, model, and key specifications. That was a useful first step. The AI Scanner is the next step: it analyzes not just what the equipment is, but what you are seeing in the photo that matters for service. This post is a preview of what it will do, how it works, and what we are still building.
The gap between equipment identification and equipment diagnosis is where experienced pool technicians earn their reputation. Any technician can read a pump nameplate. An experienced technician looks at the same pump and notices that the motor end cap has rust patterns suggesting bearing moisture ingress, that the pump pot lid has calcium deposits indicating the o-ring is not sealing completely, and that the impeller housing has cavitation marks suggesting airlock events.
That visual analysis is pattern recognition - the accumulation of years of looking at equipment in various states of condition. The AI Scanner is being built to deliver that pattern recognition from a photo, to any technician, on any service visit, regardless of their experience level with that specific equipment type.
Heat exchanger condition is one of the most consequential things to assess visually on a pool service visit. Calcium scaling on the heat exchanger face reduces efficiency and eventually causes failure. The AI Scanner will identify:
Pump and motor condition indicators that the AI Scanner will detect:
Salt cell condition assessment is particularly valuable because the visual indicators of plate degradation are well-defined but require knowing what to look for:
The AI Scanner is designed to surface observations for technician review - not to replace professional judgment. It will show you what it sees and flag what might be worth attention. The technician makes every decision. Think of it as a second set of experienced eyes that reviews your equipment photos and adds notes to your service log.
The workflow we are designing:
The output can become part of the service record - a cleaner photo-and-note trail showing what was visible, what still needs proof, and what should be verified before anyone orders or escalates.
Building a visual AI that accurately identifies equipment condition across hundreds of pool equipment brands, dozens of failure modes, and the wide range of real-world photo quality from field service visits requires significant training data. This is the technical challenge we are actively solving.
Our early access testing program serves a dual purpose: it gives professional technicians access to pre-release features, and the annotated photos contributed by testers (with consent) train the model on real field conditions. Equipment in good condition in a lab photo looks different from the same equipment that has been running in a Florida backyard for seven years with marginal water chemistry. The model needs to see the real-world version to identify it accurately.
Our architectural goal is on-device AI processing - running the scanner model locally on the phone without any internet connection required. This is technically demanding because AI vision models are large, and phone hardware has limits. Current smartphone hardware (2024+ flagships) is capable of running smaller vision models locally via Apple's Core ML and Android's ML Kit frameworks.
We are evaluating model architectures that balance accuracy against device processing requirements. If full offline processing is not achievable for the initial release, we will build a queued sync model: photos are captured offline, analyzed when connectivity is available, and results returned to the local service log automatically.
We are collecting field feedback from professional pool service technicians. If you want SplashLens to learn from a real miss, open the app after a lookup or scan and submit what was wrong, missing, or surprisingly useful.
Open SplashLens and use manual lookup, calculators, guides, and checklists right away. Create a free field profile when you want monthly AI scans, PartSnap clues, and saved proof.
Open SplashLens Free ->PartSnap focuses on possible part and equipment-family clues. The wider scanner workflow can also help with display/code reads, test-strip triage, and proof prompts. It should be treated as a cautious reference path, not a diagnosis or fitment guarantee.
Yes. Online scanner workflows are live. Manual lookup starts without an account; AI scans require a free field profile, and SplashLens Pro adds unlimited web scanner use.
AI is most useful when the photo shows labels, markings, model plates, dimensions, and clear context. SplashLens intentionally returns cautious clues and missing-proof prompts so a technician can verify against manuals, diagrams, and qualified judgment.
Our target is full offline capability using on-device processing. If offline AI processing is not achievable for first release, we will build a sync-when-connected model where photos are analyzed when the device reconnects, and results are returned to the local app automatically.