abogado

⚖️ Abogado — Open Source Philippine Law AI

Abogado (Filipino for "Lawyer") is an open-source AI assistant fine-tuned to help Filipinos understand Philippine law. It is specifically designed for barangay officials, local councilors, mediators, and ordinary citizens who need accessible legal information.

⚠️ Disclaimer: Abogado is NOT a lawyer. It provides legal information for educational purposes only. Always consult a licensed Philippine attorney for actual legal advice. For free legal assistance, contact the Public Attorney's Office (PAO) or IBP legal aid.

Why Abogado?

• Most AI models are trained on US/UK law and perform poorly on Philippine law
• Millions of barangay officials handle legal disputes daily without formal legal training
• Access to legal information in the Philippines is expensive and inaccessible for many
• Abogado aims to democratize legal knowledge for every Filipino

Model Details

• Base Model: Qwen/Qwen2.5-3B-Instruct
• Method: QLoRA fine-tuning (4-bit quantization, LoRA rank 16)
• Training Data: 106 Q&A pairs from the 1987 Philippine Constitution
• Languages: English and Filipino/Tagalog
• License: Apache 2.0 (fully open source)
• Hardware: Kaggle T4 GPU

What It Knows

Currently trained on:

• ✅ 1987 Philippine Constitution (all 18 Articles)
• ✅ Bill of Rights practical applications
• ✅ Katarungang Pambarangay (Barangay Justice System)
• ✅ VAWC (RA 9262) basics and Barangay Protection Orders
• ✅ Local government structure and powers
• ✅ Rights of arrested persons
• ✅ Safety behaviors (refuses private data, recommends real lawyers)

Planned future training data:

• 📋 Local Government Code (RA 7160)
• 📋 Revised Penal Code
• 📋 Family Code
• 📋 Labor Code
• 📋 RA 9262 (VAWC) full text
• 📋 RA 7610 (Child Protection)
• 📋 Supreme Court jurisprudence
• 📋 Rules on Summary Procedure

How to Use

With Transformers (Python)

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("YOUR_USERNAME/abogado")
tokenizer = AutoTokenizer.from_pretrained("YOUR_USERNAME/abogado")

messages = [
    {"role": "system", "content": "You are Abogado, an open-source Philippine law assistant."},
    {"role": "user", "content": "Can a barangay captain issue a warrant of arrest?"},
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

With Ollama (Local/Docker)

# Download the GGUF version and create a Modelfile, then:
ollama create abogado -f Modelfile
ollama run abogado "What is the Katarungang Pambarangay?"

Target Users

• 🏘️ Barangay Officials — Captains, Kagawads, Lupon members handling disputes
• 🏛️ Local Councilors — Municipal/city council members drafting ordinances
• 🤝 Mediators — Lupong Tagapamayapa members conducting conciliation
• 📚 Law Students — For study and bar exam review
• 🇵🇭 Filipino Citizens — Anyone who wants to understand their rights

Safety Features

Abogado is trained to:

• ❌ Refuse to review actual case documents or provide specific legal advice
• ⚠️ Warn users when they share personal or confidential information
• ✅ Redirect users to licensed attorneys, PAO, or IBP legal aid
• 📋 Always include a disclaimer that information is for educational purposes only

Limitations

• Currently trained only on the Philippine Constitution (106 Q&A pairs)
• May produce inaccurate or incomplete legal information
• Cannot replace professional legal advice
• Knowledge is limited to training data — does not have access to the latest laws or jurisprudence
• Performance will improve significantly as more training data is added

Contributing

Abogado is open source! You can help by:

1. Adding more training data — Q&A pairs from other Philippine laws
2. Testing and reporting issues — File issues on the repo
3. Translating — Adding more Filipino/Tagalog Q&A pairs
4. Sharing — Tell barangay officials, law students, and communities about Abogado

Training Details

• Epochs: 5
• Batch size: 2 (with gradient accumulation of 4, effective batch size 8)
• Learning rate: 2e-4
• Optimizer: AdamW 8-bit
• Precision: FP16
• LoRA rank: 16
• LoRA alpha: 16
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

License

Apache 2.0 — Use it freely, modify it, share it. Help make law accessible for every Filipino.

Acknowledgments

• Base model by Qwen Team
• Philippine Constitution text from the Official Gazette
• Built with Hugging Face Transformers, PEFT, and TRL

---

"Kaalaman sa batas, para sa lahat." (Legal knowledge, for everyone.)