YossoufBouzir/made-in-rwanda-recommender

# Made in Rwanda Content Recommender **AIMS KTT Hackathon · Challenge S2.T1.3** **Author:** Youssouf Bouzir **Email:** bouzir.youssouf@students.jkuat.ac.ke **Date:** 2026-04-22 --- ## Overview A CPU-friendly content-based recommender for Made-in-Rwanda products. The system uses TF-IDF over product metadata, cosine similarity for retrieval, a local-boost rule that prioritises Rwandan-made items, and a curated fallback when no strong local match is found. **Final metrics on the provided query set:** - NDCG@5: _(0.0808 0.0808 (expected 0.0808))_ - Local-presence rate (top-3): _(1.0000 (expected 1.0000 = 100.0%))_ - Curated-fallback rate : 0.0000 (expected 0.0000) **4-minute demo video:** _(............................)_ --- ## Repository structure ``` made-in-rwanda-recommender/ ├── recommender.py # main retrieval pipeline + CLI ├── eval.ipynb # evaluation notebook (NDCG@5 + local-presence rate) ├── dispatcher.md # product & business adaptation artifact ├── process_log.md # timeline + declared LLM/tool use ├── SIGNED.md # signed honor code ├── README.md # this file ├── requirements.txt # Python dependencies ├── LICENSE # MIT ├── data/ │ ├── catalog.csv # product catalog (400 SKUs) │ ├── queries.csv # evaluation queries │ └── click_log.csv # user click events └── generator/ └── synthetic_generator.py # script to regenerate synthetic data ``` --- ## Quick start (local) ```bash # 1. Install dependencies pip install -r requirements.txt # 2. Run a query python recommender.py --q "leather boots" # 3. French / code-switched query (required demo) python recommender.py --q "cadeau en cuir pour femme" --top_k 5 ``` ## Quick start (Google Colab) Open a new Colab notebook and run: ```python # Cell 1 !git clone github.com %cd made-in-rwanda-recommender !pip in …