Try My StyleFlorian Hurel Hair Couture

The Florian Hurel Hair Couture AI Consultation Platform (Try My Style) is a full-stack, in-salon AI experience for personalised skin and scalp diagnostics at point-of-service. It replaces subjective consultations with computer-vision analysis, structured AI inference, and hyper-personalised product recommendations on a premium kiosk-grade touchscreen across Florian Hurel's salon network in India.

This narrative documents product intent, architecture, engineering decisions, operational workflows, and the roadmap to scale the platform to enterprise grade.

Premium salons often face a gap between client aspiration and consultant capability:

• Skin and scalp assessments are subjective and consultant-dependent
• Recommendations lack scientific grounding visible to the client
• Diagnostic data is not captured or reused across visits
• No unified interface for consultation quality or product performance across the network
• Post-consultation engagement is weak without a report or follow-up mechanism

Try My Style addresses these gaps with one AI-driven consultation layer inside each salon.

The platform is designed to deliver:

• Clinically credible, camera-based skin and scalp analysis in under two minutes
• AI-personalised product and treatment recommendations grounded in diagnostic data
• Branded, shareable PDF reports via WhatsApp as a post-visit touchpoint
• Premium visual AI overlays that make insights tangible
• A centralised superadmin analytics layer across all salon locations

Skin analysis, scalp analysis, AI Look Advisor, Scalp View (microscope), and superadmin analytics—unified in one kiosk application.

Ten-step lifecycle from kiosk intake to WhatsApp report.

The skin visualisation step is the most distinctive product surface—one generative image pass aligned with structured scores.

• Single gpt-image-2 call produces a full 8-card diagnostic overlay (1024×1536)
• Client face photo is the edit input — identity preserved across cards
• Colour-coded overlays per concern (e.g. oiliness, pores, redness) — diagnostic, not beautification
• Generation starts at scan start, parallel to analysis API — cuts perceived wait
• quality="medium" balances ~30–40s generation vs fidelity
• UploadThing hosts the result; URL returned to the kiosk for display

Business impact

• Premium “wow” moment that justifies in-salon AI positioning
• Abstract scores become tangible and credible on screen
• Clear differentiation from standard salon consultations

• Parallel AI: gpt-image-2 visualisation + gpt-4o-mini analysis — no strictly sequential 50s wait
• React temporal dead zone: useCallback dependency ordering fixed for stable scan flow
• Axios 180s timeout on visualisation endpoint vs 60s global default
• Broken-pipe / long-poll UX: heavy work starts early so the client is not blocked on one slow hop
• Client-side React-PDF while user moves through post-result steps
• Dual cameras: normal vs microscope device IDs in localStorage + Settings UI
• Salon-scoped isolation: every request carries Bearer token; ORM queries scoped to salon
• Upload-before-analyse: face image to UploadThing; URL persisted for reports and audit
• JPEG → RGBA PNG normalisation before gpt-image-2 edit (SDK requirement)

Implemented

• JWT-based API authentication per salon device
• Salon-level isolation enforced at the ORM query layer
• Superadmin gated behind separate JWT + salon whitelist (9 access codes)
• Session analytics decoupled from core analysis (non-blocking)
• UploadThing for media persistence — no local disk dependency on kiosks

Recommended hardening

• API rate limiting per salon token
• Upload validation (type, size, malware scanning)
• Secret rotation for OPENAI_API_KEY and UPLOADTHING_SECRET
• OpenAI spend alerts and per-endpoint usage tracking
• Audit log for lead capture and WhatsApp report sends
• Retry / DLQ for failed WhatsApp deliveries

• Stateless Django — scale horizontally behind a load balancer
• UploadThing absorbs media load; no filesystem coupling
• OpenAI calls async from the user’s perspective on the frontend
• Admin analytics pre-filtered by allowed salon ID set
• Client-side PDF — zero marginal server cost per report
• ALLOWED_SALON_CODES in code — expand per deployment safely
• Future: Celery for AI jobs, Redis catalog cache, WebSocket progress, per-salon cost dashboards

Indexing priorities

• Session `created_at` for date-range analytics
• `salon_id` FK on all analysis records for isolation
• `session_completed` for funnel and completion metrics

Current setup

Django runserver for local/staging; Next.js on kiosk devices (Electron wrapper or browser kiosk mode). Environment variables include OPENAI_API_KEY, UPLOADTHING_SECRET, and SECRET_KEY.

Recommended production setup

• Backend: Gunicorn + Nginx on VPS or containers (WSGI-compatible)
• Frontend: Next.js standalone build deployed as PWA or Electron
• Database: PostgreSQL (migrate from SQLite for production)
• Media: UploadThing (already CDN-backed)
• CI/CD: GitHub Actions — backend deploy + Next.js artifact per release

Monitoring targets

• OpenAI API error rate and latency per endpoint
• UploadThing upload success rate
• WhatsApp delivery success rate
• Session completion funnel drop-offs

Example statements for interviews and portfolio accuracy.

Production systems as icon clusters—hover or focus each tile for the technology name.

Quantified outcomes matter for investors and enterprise buyers—replace template lines with live analytics when available.

Metrics to instrument

• Daily skin and scalp analyses per salon
• Full-flow completion rate (camera → lead capture)
• Lead capture conversion rate
• WhatsApp report delivery success rate
• AI visualisation generation time (target < 40s)
• Product recommendation engagement (cart adds)
• Repeat sessions by phone number
• Expert tips satisfaction ratings

Impact statement templates

• “Parallel AI generation reduced perceived wait from ~50s to ~35s per consultation.”
• “AI-ranked recommendations increased average products viewed per session from N to M.”
• “WhatsApp PDF reports maintained X% open rate as a post-visit retention touchpoint.”
• “Skin visualisation step improved consultation satisfaction from A to B stars.”

Reserve frames for architecture, flows, kiosk UI, admin analytics, PDF samples, and expert tips. Export at 2× for crisp portfolios.

Live footprint

Deployed across nine Florian Hurel Hair Couture locations in India—embedded as a core in-salon service, not a standalone consumer app.

• Walkeshwar, Bandra, Waterfield (×2), Juhu, Mumbai
• Pune
• Ahmedabad — Sindhubhavan & Ambli
• Hyderabad — Jubilee Hills

Validation indicators

• Multi-location deployment across nine salons signals enterprise readiness
• Skin + scalp modules prove extensible dual-service architecture
• WhatsApp PDF creates measurable post-visit brand touchpoint
• Superadmin analytics shows network-level operational maturity

Try My Style is not a single feature—it is a coordinated AI product ecosystem spanning kiosk UX, computer vision, generative visualisation, personalised recommendations, automated report delivery, and network-wide analytics. Parallel generation, structured outputs, and client-side PDF are engineering-grade choices that support Florian Hurel's premium positioning.