Strategic Framing of an AI Project: Clothing Recommendation by Photo

Cadrage IA, Computer Vision, System Design, Azure, MLOps, ROI, RGPD, Product Management

AI Project Framing: Clothing Recommendation Application (POC)

Overview

This project implements the entire framing cycle of an AI initiative, from translating a business need into a technical approach to presenting a costed proposal to an Executive Committee. The objective is to transform a product idea — a mobile application for recommending clothing items from a photo — into a credible, budgeted, and deployable Proof of Concept.

The project applies best practices for AI project framing to ensure the technical feasibility, economic viability, and regulatory compliance of a solution before any development investment.

Objectives

• Translate a business need into a concrete and demonstrable AI/ML approach.
• Define measurable and relevant success criteria from a business perspective.
• Design an end-to-end System Design for the target production solution.
• Distribute roles and responsibilities within a Data team.
• Build a delivery timeline and a staffing plan.
• Cost out (one-shot and recurring) expenses and project the ROI.
• Identify RGPD risks and propose mitigation solutions.

Demonstrated Skills

Business need → AI approach translation

• Analysis of a statement of need (product sponsor email) and reformulation as an ML problem.
• Reasoned comparison of two technical approaches: visual similarity search (embeddings) vs multimodal LLM analysis.
• Selection of candidate datasets and framing of the analysis scope for a quickly demonstrable POC.
• Definition of both technical and business success criteria.

System Design & Cloud Architecture

• Design of an end-to-end architecture, from user data (photo) to raw data.
• Mapping of each technical component to an equivalent Microsoft Azure service.
• Anticipation of solution scalability based on marketing traffic projections.

Data Team Planning & Management

• Distribution of roles among 3 Data Scientists, 2 Data Engineers, 1 Tech Lead, and 1 MLOps for each technical component.
• Construction of a phase-based timeline with staffing rates per profile.
• Definition of governance instances (Steering Committee with business experts) to ensure the continuous relevance of the solution.

Economic Costing & ROI

• Estimation of person-day costs per phase and per profile.
• Distinction between one-shot costs (development) and recurring costs (compute, storage).
• Construction of best-case / worst-case scenarios based on conservative assumptions.
• Calculation and graphical visualization of the break-even point, based on sales gains estimated by marketing.

RGPD Compliance & Risk Management

• Mapping of personal data processed, beyond just image analysis.
• Translation of key RGPD principles into concrete risks for the project.
• Identification of anonymization points in the System Design.
• Guarantee of non-reuse of user data for LLM training and proposal of alternatives to public APIs.

Architecture

                Utilisateur (photo)
                       │
                       ▼
              Application mobile
                       │
                       ▼
         Ingestion & anonymisation
                       │
                       ▼
         Stockage sécurisé (Azure)
                       │
        ┌──────────────┴──────────────┐
        ▼                             ▼
  Extraction d'embeddings      LLM multimodal
  (Computer Vision)             (alternative)
        │                             │
        └──────────────┬──────────────┘
                        ▼
              Moteur de recommandation
                        │
                        ▼
              Résultats personnalisés
                        │
                        ▼
                Application mobile
                        │
                        ▼
           Monitoring & feedback métier

Technical & Methodological Stack

AI / Machine Learning

• Visual similarity approaches (image embeddings)
• Multimodal LLMs (image analysis)
• Google Dataset Search (data sourcing)

Cloud

• Microsoft Azure (managed services)
• Azure Pricing Calculator

Project Management & Framing

• System Design
• RACI / responsibility distribution
• Timeline & phase-based staffing

Compliance

• RGPD (principles and application)
• Personal data anonymization

Deliverable

• Formalized PowerPoint (.pptx) presentation for an Executive Committee

Applied Framing Best Practices

• Scope tightened to the critical AI component (no over-engineering of the POC)
• Business-oriented success criteria
• Conservative cost assumptions
• Consideration of scalability from the design phase
• Anticipation of regulatory risks upstream of development
• Communication adapted to non-technical stakeholders (Executive Committee, DPO)

Results

At the end of this project, the framing document allows for:

• demonstrating the added value of AI for the chosen use case;
• validating the technical feasibility of the POC in the short term;
• justifying a precise budget and resource allocation;
• projecting a break-even date for the project;
• anticipating RGPD risks before any development;
• convincing a demanding Executive Committee (VP Product, Chief Data Officer, DPO).

Acquired AI Framing Skills

• End-to-end AI project framing and structuring
• Translation of a business need into a technical solution
• Cloud System Design (Azure) conception
• Budget estimation and ROI calculation
• Data team planning and staffing
• Data governance and RGPD compliance
• Communication and presentation to management bodies