Contact person: AI Toilet Cleaning Assistant App

Phone:Show

Email:Show

Location: Гродно, Belarus

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"I’m assembling the first-generation, AI-powered toilet-cleaning assistant. The goal is an end-to-end MVP that lets a user open a mobile app, snap or upload a toilet photo, receive real-time waste detection overlays, and instantly see tailored cleaning advice. The core logic already lives in a legacy Pascal codebase, orchestrated by our in-house X-ui-script workflow engine, so every new module has to speak their language while adding modern AI muscle.

How the pieces fit
• AI layer – a Python (TensorFlow/PyTorch) model must accept a JPEG/PNG, return pixel-accurate masks and human-readable “waste type” labels. Pack it behind a lightweight gRPC or REST endpoint so Pascal can call it.
• Backend integration – extend the existing Pascal services, expose new endpoints, and trigger X-ui-script sanitation routines once the AI responds. C++ is welcome for any performance-critical image ops.
• Mobile experience – iOS + Android app is the primary interface. It needs three live features: camera/gallery upload, real-time waste detection overlay, and a cleanly formatted recommendation panel. React Native, Flutter, or a thin JavaScript/TypeScript SPA wrapped with Capacitor are all acceptable as long as performance and camera access stay smooth.
• Desktop utilities – CLI or lightweight GUI test harnesses that run identically on Windows, macOS, and Linux so I can batch-verify the AI and sanitation scripts outside the mobile flow.

Acceptance checklist
– Photo passes from mobile to backend in <2 s over Wi-Fi.
– Detection overlay aligns within ±3 px on a 1080 p frame.
– Recommendation text pulls from the X-ui-script rule set and renders under 200 ms.
– All services containerised, documented, and runnable with one docker-compose up.
– Source code builds cleanly on the three desktop OS targets and both mobile platforms.

Once the MVP proves stable, we’ll scale the model, add user profiles, and hook in subscription billing, so clean, modular code is essential from day one. If you’ve balanced Pascal oddities with modern AI stacks before, this should be a fun sprint." (client-provided description)