Methodology
How Lawyerly Works
A verified knowledge graph supplies the law — AI only supplies the language.
Lawyerly is grounded in a verified, hand-built knowledge graph of Pakistani statutes and case law. AI models are used for language generation and retrieval — they are not the source of legal truth. Citation and case-status checks come from our own database, not from the AI model itself.
Which models we use
| Feature | Model | Hosting Country |
|---|---|---|
| Legal research chat | gpt-5.6-luna gpt-5.4-nano (fallback) | United States (OpenAI) United States (OpenAI) |
| Document & argument drafting | gpt-5.6-luna gpt-5.4-nano (fallback) | United States (OpenAI) United States (OpenAI) |
| Reading judgments you upload | gemini-flash-latest | United States (Google) |
| Reading an advocate licence card you upload | gpt-5.6-luna | United States (OpenAI) |
| Community verification of case treatments | deepseek-v4-flash gpt-5.6-luna (second check) | China (DeepSeek) United States (OpenAI) |
Citations are grounded in Lawyerly’s own knowledge graph rather than the model’s memory: the drafting and chat text is generated by a third-party model. In drafts, every citation is checked against Lawyerly’s database; in chat, citations are checked against it too, and anything that cannot be confirmed is flagged in the text rather than passed off as verified. Overruled and distinguished precedents are flagged with the judgment that did it.
Have questions or concerns about the models we use or how your data is handled? Contact us at support@startuplab24.com.