muma005/shamba_qi

# shamba_qi # ShambaQA: Swahili Agricultural Pest & Disease Advisory QA Dataset > **Zero Swahili agricultural QA datasets exist. This fills that gap.** ShambaQA is an open-source question-answering dataset in Swahili (Kiswahili) covering crop pest and disease diagnostics for East African smallholder farmers. It pairs realistic farmer questions with authoritative expert answers sourced from agricultural extension materials. --- ## Quick Start ```python from datasets import load_dataset ds = load_dataset("username/shambaqa") print(ds["train"][0]) ``` --- ## Key Statistics | Metric | Value | |--------|-------| | Total QA pairs | *[UPDATE AFTER BUILD]* | | Language | Swahili (Kiswahili) | | Dialect variants | Kenyan Swahili, Tanzanian Swahili, Standard | | Crops covered | 12 | | Pest/disease entries | ~150 | | Categories | 11 | | Format | JSONL + CSV | | Train / Dev / Test | 80% / 10% / 10% (stratified) | --- ## Scarcity Evidence As of April 2026, searching HuggingFace for "Swahili agriculture", "Swahili pest", and "Kiswahili QA" returns zero matching datasets. Screenshots documenting this search are included in `docs/`. *[INSERT SCREENSHOTS HERE]* --- ## Why This Dataset 33 million+ smallholder farmers in East Africa lose 30–40% of crops annually to pests and diseases (FAO). Digital advisory tools like iShamba and Plantwise are actively deploying SMS/WhatsApp-based farmer support but lack the NLP backbone for Swahili. ShambaQA provides the training and evaluation data to build that backbone. --- ## Target Users - NLP researchers working on low-resource language QA - Agritech startups building farmer advisory chatbots - Agricultural extension organizations digitizing outreach - Multilingual LLM evaluation teams needing domain-specific Swahili benchmarks --- ## Dataset Fields See DATA_DICTIONARY.md for complete column definitions. Key fields: `question_sw`, `answer_sw`, `crop`, `pest_disease`, `category`, `severity`, `source_ref`, `question_source` …