AI-Based Mobile App for Image Recognition (AI PoC Development)

About the Client

Our client is a premier manufacturer of professional cleaning chemicals, floor care chemicals, disinfectants, and sanitary maintenance brands. Founded in 1920 and family-owned, the client has a rich history of innovating cleaning products and solutions for a wide range of industries and markets, including building and facilities maintenance, janitorial cleaning, industrial cleaning, food processing, healthcare, and education.

Challenge

With a century of experience in chemical formulation, packaging, and private chemical branding, the client manufactures cleaning chemicals for nearly every application. They needed an AI-powered application for distributors to sell their cleaning products.

The challenge was to build a system that identifies surfaces requiring cleaning (e.g., floors, tables, sinks), recommends the appropriate methods and products from the client’s range, and validates the concept with an interactive Figma prototype for a conference showcase.

Project Scope

Our team developed an AI-powered application that recognizes surfaces and objects to recommend the most appropriate cleaning products and procedures based on material and context. From the main screen, users can capture a photo with the camera, select one from the gallery, or upload a file. The app then accurately classifies the item and provides step-by-step cleaning guidance tailored to the recognized surface.

The project consisted of 4 phases: (0) Concept & UX Validation, (1) Proof of Concept, (2) MVP and App Store Release, (3) Model Ownership and Improvement – in progress.

Solution We Delivered

Prototype (Conference Demo)

Designed core user journeys (capture → recognize → recommend → next steps) and built a clickable Figma prototype.
Demonstrated the concept live at a conference; won an innovation award, validating user interest and business viability.
Captured early feedback on UX, copy, and recommendation presentation to inform PoC scope.

Proof of Concept

Shipped a working mobile app with an OpenAI-powered classification backend and lightweight admin tools for labels/product mapping.
Tested against a predefined set of 50 surface types; instrumented analytics to track accuracy, latency, and user flows.
Ran a closed beta to validate core functionality, gather qualitative feedback, and prioritize improvements (e.g., edge cases, low-light capture).
Established baseline metrics (top-1/top-3 accuracy, time-to-result) and acceptance thresholds for production.

Production Release and Future Enhancement

Published to the app store with privacy, consent, and telemetry in place; added crash monitoring and performance budgets.

The following steps are to switch to a client-owned custom ML model, enhance in-domain accuracy, and improve confidence calibration. Also, introduce MLOps capabilities (versioning, evaluation sets, staged rollout/rollback, drift monitoring) to enable safe, iterative model updates.

Client Results

Our client got a full-stack AI solution that evolved from an award-winning prototype into a production-grade application, leveraging both generative AI and proprietary ML models. After testing iOS devices, the client presented the concept at an industry event, where it gained substantial traction and opened doors to key stakeholders.

Have a Question? Let’s Get in Touch!

Tell us what you’re building or where you’re stuck. We work with engineering and product teams on custom software, AI & ML, cloud infrastructure, DevOps, and UI/UX — from early scoping to long-term delivery. One conversation is usually enough to know whether we’re the right fit.