This mini project is an AI-powered recommender system that matches a user’s vibe to a set of fashion products.
You can type something like “energetic urban chic,” and it will find products that best fit that mood — for example, sporty sneakers or a leather jacket.
The goal was to explore how embeddings and cosine similarity can enable intelligent, context-aware recommendations.
Although the original plan was to use the OpenAI Embeddings API, fake embeddings were generated with NumPy due to quota limits — the logic remains the same.
- A small dataset of 10 mock fashion products is created using Pandas.
- Each product includes a short description and vibe tags (e.g., “boho”, “chic”, “cozy”).
- Product descriptions and user vibe queries are converted into 1536-dimensional vectors (simulated embeddings).
- Cosine similarity is computed between the query and all products.
- The top 3 most similar products are displayed along with their similarity scores.
- Multiple vibe queries are tested, and query latency is measured and visualized.
- Python
- Pandas
- NumPy
- scikit-learn
- Matplotlib
- (Planned) OpenAI Embeddings API
This project demonstrated how recommendation systems can use vector similarity to find contextually related items.
Even with simulated embeddings, it shows the underlying mechanism clearly.
Next steps:
- Integrate real embeddings via OpenAI or similar models.
- Add FAISS or Pinecone for scalable vector search.
- Build a minimal UI for interactive vibe matching.
AI at Nexora, to me, means building technology that feels personal and intuitive.
The “Vibe Matcher” project was a small experiment to show how AI can interpret moods and creativity, not just data or metrics.
It blends logic with emotion — exactly the kind of AI experience I want to help create.