DrUkachi/ktt-made-in-rwanda

Content-based niche-first recommender that surfaces Made-in-Rwanda artisan products over global brands. Multilingual TF-IDF (EN/FR/Kinyarwanda) + local-boost + fairness cap. <5 ms per query on CPU. Includes a low-tech weekly lead workflow for offline artisans. # S2.T1.3 — 'Made in Rwanda' Content Recommender **AIMS KTT Hackathon · Tier 1 · RetailTech · Recommender Systems · Low-Tech Distribution** > A content-based 'niche-first' recommender that nudges buyers toward local Rwandan > artisan products — with multilingual query support (English, French, Kinyarwanda) > and a low-tech weekly lead delivery workflow for offline artisans. --- ## The Problem Generic e-commerce recommenders rank by global popularity, polished marketing copy, and SEO authority. On a typical search engine, a query for "leather boots" returns Timberland and Dr Martens — not the leatherworker in Nyamirambo who has no website, no smartphone, and writes product descriptions on a notebook in Kinyarwanda. This recommender flips the priority. It treats the catalog of Made-in-Rwanda artisan products as the primary inventory, mixes in a realistic competitive set of international brands (Pandora, Coach, Timberland, IKEA, …) so the local-boost is actually doing real work, and then enforces three rules: 1. A Made-in-Rwanda product **always** appears at rank 1. 2. No single artisan dominates more than 15% of the top-10 (fairness cap). 3. Multilingual queries — including French and Kinyarwanda code-switching — work without a translation API. The output of the search engine is then converted into weekly lead packets that can be **delivered via SMS or a field-agent voice call** to artisans without internet access. See `dispatcher.md` for the full distribution design. --- ## Quick Start **Two commands — works on Google Colab (free CPU) or any Python 3.10+ environment:** ```bash pip install -r requirements.txt python recommender.py --q "leather boots" ``` No model downloads. No GPU. Cold start under 1 second. ### Example queries ```bash # English python recommender.py --q "leather boots" # French (FR→EN dictionary normalisation built-in) python recommender.py --q "cadeau en cuir pour femme" # Code-switched (Kinyarwanda + English + French) python recomm …