siltanukifilie/aims-ktt-day2-g2-t2-2-grant-matcher

Multilingual Grant & Tender Matcher for African entrepreneurs — ranks relevant grants/tenders by business profile and generates plain-language summaries in English and French. Built for AIMS KTT Fellowship Hackathon 2026 · Day 2 · Challenge T2.2 # aims-ktt-day2-g2-t2-2-grant-matcher # 🚀 Multilingual Grant & Tender Matcher (AIMS T2.2) ## 📌 Overview This project builds a lightweight system that matches businesses and cooperatives with relevant funding opportunities (grants and tenders). It is designed for real-world deployment in low-resource environments, supporting: - 🌍 Multilingual documents (English & French) - ⚡ Fast retrieval using BM25 + TF-IDF - 🧠 Profile-based matching - 📊 Measurable performance (MRR@5, Recall@5) --- ## 🎯 Problem Small businesses and cooperatives face major challenges: - Difficulty finding relevant tenders - Language barriers (English/French documents) - Information overload from unstructured sources --- ## 💡 Solution We built a system that: 1. Parses tender documents (PDF, HTML, TXT) 2. Extracts structured information (sector, region, budget, deadline) 3. Matches tenders to business profiles 4. Ranks the most relevant opportunities 5. Generates short summaries --- ## 🧠 System Pipeline Parse → Extract → Index → Match → Rank → Summarize ### Matching Logic - BM25 ranking (primary retrieval) - TF-IDF scoring (support) - Sector and region boosting - Cross-lingual normalization (FR → EN) --- ## 📊 Results | Metric | Score | |----------|--------| | MRR@5 | 0.7167 | | Recall@5 | 0.5667 | 👉 Significantly above random baseline --- ## 🧪 Evaluation Run locally: ```bash python evaluate.py ``` Or explore: - eval.ipynb for reproducible evaluation --- ## 🖥️ Live Demo 👉 Hugging Face Space huggingface.co Features: - Select business profile - View ranked tenders - Interactive UI --- ## 🧑‍💼 Product Design (Village Agent) Designed for real-world deployment: - 📱 WhatsApp audio delivery - 🌍 Multilingual support - 📡 Low-bandwidth environments - 👥 Cooperative-focused access See: village_agent.md --- ## 📁 Project Structure . ├── matcher.py ├── evaluate.py ├── data_generator.py ├── eval.ipynb ├── dataset/ ├── summaries/ …