satyajitdas/bharatschemes-v1

--- language: - en - hi license: apache-2.0 task_categories: - question-answering - text-generation tags: - india - government-schemes - welfare - csr - ngo - instruction-tuning - bharatschemes size_categories: - 1K<n<10K --- # BharatSchemes Dataset v1 **India's first comprehensive instruction-tuning dataset for government welfare schemes, state programs, and corporate CSR/NGO initiatives.** ## Overview This dataset contains 2029 instruction-following Q&A pairs covering: | Category | Coverage | |----------|----------| | Central Government Schemes | MyScheme.gov.in (3000+ schemes) | | State Government Schemes | All 28 states + 8 UTs | | Corporate CSR Programs | Tata, Infosys, Wipro, Reliance, Azim Premji, Adani, Mahindra | | NGO Initiatives | Pratham, Akshaya Patra, GiveIndia, Teach For India, CRY, Goonj | | Multilateral Programs | UNDP India | ## What Makes This Unique 1. **Persona-based queries**: "I'm a 30-year-old woman farmer in Rajasthan. What schemes can help me?" 2. **Cross-state comparisons**: "Compare health insurance schemes across states" 3. **Application guidance**: Step-by-step process for each scheme 4. **Unified coverage**: Central + State + CSR + NGO in one dataset (first of its kind) ## Format ChatML format (compatible with MLX, Axolotl, LLaMA-Factory): ```json { "messages": [ {"role": "system", "content": "You are BharatSchemes Assistant..."}, {"role": "user", "content": "What is PM-KISAN?"}, {"role": "assistant", "content": "PM-KISAN is..."} ] } ``` ## Splits - **train.jsonl**: 1927 examples - **valid.jsonl**: 102 examples ## Usage ```python from datasets import load_dataset ds = load_dataset("satyajitdas/bharatschemes-v1") ``` ## Author **Satyajit Das** (IIT Kharagpur alumnus) - Cambridge PhD in Engineering - Consensus Engineer @ Offchain Labs (Ethereum / Prysm) - Protocol Guild Member - Publication: Science (nano architecture, ~1000 citations) ## Citation ```bibtex @dataset{bharatschemes2026, title={BharatSchemes: Instruction Dataset for Indian Welfare Scheme Navigation}, author={Das, Satyajit}, year={2026}, url={https://huggingface.co/datasets/satyajitdas/bharatschemes-v1} } ``` ## License Apache 2.0. Source data from public government websites.