gabrielntwari/AIMS-KTT-Hackathon-Made-in-Rwanda-Content-Recommender

Content-based 'niche-first' recommender that nudges buyers toward Made-in-Rwanda local products, with offline SMS lead delivery for non-smartphone artisans. ## About This project is a content-based product recommender built for the AIMS KTT Hackathon (S2.T1.3 — 'Made in Rwanda' Content Recommender). Local Rwandan artisans lose sales to global e-commerce algorithms that favour high-volume international sellers. This recommender flips that dynamic — it ranks local 'Made in Rwanda' products first, handles multilingual queries (English, French, Kinyarwanda), corrects misspellings, and delivers weekly lead summaries to artisans who have no smartphone via SMS. ### Key Features - Hybrid TF-IDF index (word + character n-grams) for typo robustness - FR/Kinyarwanda → English query translation - Local-boost: all Rwandan products scored higher, niche districts get an extra 10% - Fairness cap: no single artisan occupies more than 15% of top recommendations - Fallback curation when no strong local match exists - CLI interface: `python Recommender.py --q "leather boots"` - Offline artisan workflow: weekly SMS digest delivered via cooperative agent #### Video Link drive.google.com ### Performance | Metric | Score | |---|---| | NDCG@5 | 0.93 | | % queries with target in top 5 | more than 90% | | Avg rank of target product | 1.3 | | Query languages supported | English, French, Kinyarwanda | ### Quick Start ```bash git clone github.com cd made-in-rwanda-recommender pip install -r requirements.txt python generator.py python Recommender.py --q "leather boots" python Recommender.py --q "inkweto" python Recommender.py --q "cadeau en cuir pour femme" ```