OndiekiFrank/kenya-constitution-ai-agent

An NLP-powered AI assistant for exploring the Kenyan Constitution. Uses Retrieval-Augmented Generation (RAG) with NLP pipelines for accurate, accessible legal Q&A. # kenya-constitution-ai-agent An NLP-powered multilingual AI assistant for exploring the Kenyan Constitution in English and Swahili. Uses Retrieval-Augmented Generation (RAG) with NLP pipelines for accurate, accessible legal Q&A. ## Table of contents - Business Context - Datasets Used - Directory Structure - Data Preparation - Exploratory Data Analysis - Visualizations - Models and Performance - Model Interpretability - Findings - Recommendations - Conclusion - Team - License ## Business Context Kenya’s Constitution is published in both English and Kiswahili, but citizens, students, and policymakers often face challenges in accessing and understanding it. This project aims to: - Classify legal text by language. - Retrieve relevant constitutional articles. - Support *question answering* and knowledge democratization. ## Dataset used All data files are located in the Data/ folder: | File Name | Source | |-----------|--------| | The_Constitution_of_Kenya_2010.pdf | Kenya Law (Official English Version) | | Kielelezo_Pantanifu_cha_Katiba_ya_Kenya.pdf | Kenya Law (Official Kiswahili Version) | > All datasets were obtained from publicly available legal sources and are used *strictly for educational and research purposes*. ## Directory Structure kenya-constitution-ai/ ├── Data/ │ ├── The_Constitution_of_Kenya_2010.pdf │ ├── Kielelezo_Pantanifu_cha_Katiba_ya_Kenya.pdf │ ├── kenya_constitution_structured.csv │ └── kenya_constitution_prepared.csv ├── Notebooks/ │ └── Kenya_Constitution_AI_Agent.ipynb ├── Models/ │ ├── traditional_ml_models.pkl │ └── deep_learning_model.h5 ├── Images/ │ └── (Saved charts, dashboards, and figures) ├── README.md └── .gitignore ## Data Preparation - *Dataset size*: 121 legal text samples - *Languages*: English & Kiswahili - *Preprocessing steps*: - Tokenization - Stopword removal - Vectorization (TF-IDF, embeddings for DL) - *Splits*: Train (80%) / Test (20%) ## Exploratory Data Analysis In this section we explo …