BharatSchemes-7B

India's first fine-tuned LLM for navigating government welfare schemes, state programs, and corporate CSR/NGO initiatives.

Helping 1.4 billion citizens discover the support they're entitled to.

Model Description

BharatSchemes-7B is a Qwen 2.5 7B Instruct model fine-tuned with LoRA on the BharatSchemes dataset — a comprehensive collection of Indian central government schemes, state programs, corporate CSR initiatives, and NGO programs.

What it can do

• Eligibility matching: "I'm a farmer in Bihar with 2 acres. What schemes help me?"
• Scheme explanation: "Tell me about Chiranjeevi Yojana"
• Application guidance: "How do I apply for PM-KISAN?"
• Cross-state comparison: "Compare women's empowerment schemes across states"
• CSR/NGO navigation: "What does Tata Trusts offer for education?"

Training

• Base model: Qwen/Qwen2.5-7B-Instruct
• Method: LoRA (rank 16, alpha 32)
• Framework: MLX (Apple Silicon optimized)
• Hardware: Apple M4 Max (96GB unified memory)
• Dataset: satyajitdas/bharatschemes-v1

Usage (MLX)

from mlx_lm import load, generate

model, tokenizer = load("satyajitdas/BharatSchemes-7B")
prompt = "I'm a single mother in Karnataka with two school-going daughters. What government schemes can help me?"
response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
print(response)

Author

Satyajit Das (IIT Kharagpur alumnus)

• Cambridge PhD in Engineering
• Consensus Engineer @ Offchain Labs (Ethereum / Prysm)
• Protocol Guild Member
• Science publication (~1000 citations)

Limitations

• Data reflects schemes as of early 2026; schemes are updated/discontinued regularly
• Not a substitute for official verification at myscheme.gov.in
• Coverage of state schemes varies; some states have more data than others
• Primarily English; Hindi support is inherited from base model

License

Apache 2.0