Generative AI · L'Oréal Brandstorm 2026
Scent Genome Engine
A GenAI fragrance recommendation engine built for L'Oréal Brandstorm 2026: embeddings-based note similarity plus LangChain prompt chains that generate personalised scent stories.
Explainable by design every recommendation states its reasoning, with a generated scent narrative
The idea
Fragrance discovery is stuck between two bad options: overwhelming shelves and generic quizzes. For L’Oréal Brandstorm 2026 I prototyped a recommendation engine that treats scent like a genome: decompose fragrances into notes, learn similarity between them, and explain every recommendation in plain language.
What I built
- Note similarity scoring using embeddings and collaborative filtering, with botanical constraints and neuroscience-backed personalisation baked into the ranking
- A GenAI scent story generator using LangChain prompt chains over the OpenAI API, producing personalised fragrance narratives mapped to user profiles and the botanicals of their city
- Explainability by design: fine-tuned prompt chains so every recommendation says why, and renders as a shareable tile with maps and notes
- Streamlit front end on a MySQL backend so the whole prototype is something you can actually click through
Why it matters
Most recommendation engines are black boxes. This one shows its reasoning, and the generated narratives turn a model output into something a customer actually wants to share. The same pattern applies to enterprise decision support: retrieval plus generation plus explainability.