Kylapurity/Irrigation_ChatBot

Most small-scale farmers have no info on how the use of Irrigation in agriculture will increase their yields and income. Having been able to use Shamba Irrigation Bot they will use it to learn more of irrigation and i's management # Shamba IrrigationBot – AI Chatbot for Irrigation Support in Kenya *Shamba IrrigationBot* is an artificial intelligence-powered chatbot designed to assist Kenyan smallholder farmers by providing timely and accurate irrigation advice. The chatbot addresses water management challenges, promotes better crop yields, and supports food security by sharing expert knowledge through an intuitive web interface. ## Project Overview This project fine-tunes the T5-base model on a curated Water Irrigation Dataset to generate dynamic, context-aware responses to farmers’ questions. The chatbot is lightweight and accessible through a browser, making it practical for farmers in rural areas with limited internet connectivity. ## Dataset The dataset used was obtained from Hugging Face and contains 3,851 entries. Each entry includes: - *QUESTION.question*: A farmer's query (e.g., "How do I irrigate maize?") - *ANSWER*: An expert response (e.g., "Use drip irrigation to conserve water and ensure even moisture distribution.") The dataset covers: - Crop-specific irrigation - Drip systems - Water technology and pipe selection - Irrigation methods for yield improvement ## Data Preprocessing To prepare the dataset for training with the T5-base model (220M parameters), the following steps were applied: - *Dataset Split*: 80% training, 10% validation, 10% test - *Text Normalisation*: Lowercased text, removed extra spaces and special characters - *Tokenization*: Used T5-base tokenizer with a maximum sequence length of 128 tokens - *Input Format*: "generate irrigation advice: [query]" - *Quality Checks*: Removed duplicates, ensured query diversity - *Monitoring*: Used Weights & Biases (wandb.ai) to track training and hyperparameter performance ## Model Training The T5-base model was selected for its ability to generate natural language responses with moderate computational demands, making it ideal for use in limited-resource environments. ### Fine-Tuning Details | Parameter …