A store that understands itself.

The brief

LC Waikiki operates over 1,300 stores in more than 60 countries. At that scale, a percent-point gain in conversion, dwell time or stock accuracy translates into billions of lira. The R&D question was concrete: could an AI-driven, computer-vision + RFID platform give store managers a real-time view of what actually happens between the door and the fitting room?

The research had to be delivered in a form two very different audiences could evaluate at once: a technical team who would look at the algorithms and the models, and a non-technical audience who would only see the demo and the numbers.

What I built

I led the R&D workstream end-to-end: designed the algorithms, trained the vision models, and built the full interactive demo that packages the whole system as a live product. Alongside the demo I put together a written proposal covering competitive analysis, technical architecture, a phased rollout plan, hardware/software specs, KVKK compliance and the financial model.

• Vision models — trained end-to-end. Custom detection and tracking models for garment identification, pick-up events and dwell time. Data curation, labeling schema, training pipeline and evaluation loops all owned by me.
• Live camera panels that highlight products picked up, dwell time per SKU and suspicious-behaviour scenarios (theft flags on specific SKUs).
• Real product data — actual retail SKU codes and best-sellers, so the demo reads like a real store.
• Manager dashboard with an AI-driven decision chain: from a raw signal ("customer holds SKU-4388 for 47 seconds") to a business action ("initiate transfer from warehouse").
• Customer density heatmap across the sales floor, and a store-level revenue model that scales the projections from a single store to a full multi-hundred-store network.

Live demo

Analysis and results

The financial model was built from the ground up on real retail numbers. Baseline monthly revenue was set at ₺48.75M per store; the platform's projected uplift — driven by better stock decisions, dwell-time-based merchandising and loss prevention — worked out to ≈₺111M/year per store, or ≈₺11.1B/year at 100 stores, against a pilot investment budget of ~$355K.

The project is still in the R&D phase — the algorithms, models and interactive demo are complete, but the platform has not yet gone into production. Beyond the specific proposal, the exercise proved out a reusable pattern I've kept using since: package deep-tech systems as a live, product-shaped demo, not a slide deck.

Where it's going

The same platform architecture and its adaptations are currently in active conversations with several retail brands in Türkiye — rescaled financials, brand-specific palettes and category-appropriate visual identities per client. Names and details stay under NDA until the conversations mature.