stindardlogic/legal-reasoning-sft-100k

Legal Reasoning SFT (100K)

100,000 ShareGPT conversations demonstrating expert-level legal reasoning across contract analysis, M&A due diligence, IP law, regulatory compliance, employment law, and corporate transactions.

Motivation

Legal AI is one of the highest-value enterprise AI applications — law firms, in-house counsel, and legal tech platforms need models that can reason through contracts, identify risks, and explain legal concepts with the precision of a trained attorney. Models commonly fail by:

• —Surface-level analysis: Identifying clauses without explaining why they're risky or what the consequences are
• —Missing the business context: Legal advice divorced from commercial reality — telling clients what the clause says but not what it means for their business
• —One-size-fits-all answers: Ignoring jurisdiction-specific differences that completely change the legal outcome
• —Missing practical remedies: Describing problems without suggesting specific negotiating language or solutions
• —Oversimplifying complex regulatory frameworks: GDPR, HIPAA, securities law — these require nuanced treatment, not checkbox summaries

This dataset trains models to reason like a seasoned commercial attorney: identifying real risks, explaining the business consequences, and providing actionable next steps.

Dataset Description

100,000 conversations across 6 legal categories:

Category Distribution

Format

{
  "conversations": [
    {
      "from": "human",
      "value": "Review this SaaS subscription agreement clause and identify any risks for the customer..."
    },
    {
      "from": "gpt",
      "value": "## Contract Risk Analysis\n\n**Clause 1: Feature Modification Without Notice**\n\n**Risk Level: HIGH**\n..."
    }
  ],
  "metadata": {
    "category": "contract_analysis",
    "context": "SaaS subscription agreement"
  },
  "id": "abc123"
}

Key Properties of Responses

1. Risk quantification: Every identified risk is rated (LOW/MEDIUM/HIGH) and explained in business terms — not just "this is unusual" but "this creates X exposure because Y."

2. Jurisdiction awareness: Responses explicitly identify when the answer differs by jurisdiction (California vs. other states on non-competes, EU GDPR requirements, state securities laws), rather than giving US-generic answers.

3. Negotiating language provided: Contract analysis responses include specific redline language — not just "negotiate this" but the actual words to propose.

4. Business context integrated: Legal analysis connects to commercial outcomes: deal valuation impact, operational implications, investor/acquirer concerns.

5. Practical next steps: Every analysis ends with an actionable prioritized list: what to do immediately, what to negotiate, what to monitor.

6. Complexity calibrated: Simple questions get concise answers; complex regulatory analysis gets the full treatment with proper nuance.

Use Cases

• —SFT fine-tuning for legal AI platforms (Harvey, Ironclad AI, Kira, Luminance)
• —Training AI contract review tools for enterprise legal teams
• —Building AI compliance assistants for healthcare, finance, and technology companies
• —Improving model performance on legal reasoning and structured analysis
• —Training AI for M&A due diligence automation
• —Fine-tuning models for legal tech startups building document review and contract intelligence products

License

Apache 2.0