kenax69/EU-Air-Passenger-Rights-Instruction-Dataset-AAA

EU Air Passenger Rights Instruction Dataset (Production-Grade Sample)

This repository contains a premium, human-curated instruction dataset focused on European Union law (Regulation (EC) No 261/2004 and relevant Court of Justice case law).

It has been converted and unified into the production-standard ShareGPT/Alpaca conversational format, making it instantly ready for supervised fine-tuning (SFT) of open-weights LLMs (such as LLaMA 3, Mistral, or Qwen) or for indexing in advanced RAG pipelines.

🛠️ Dataset Features & Quality Standards

Most automated data dumps rely on naive scraping and lazy extraction, resulting in dirty rows, data duplication, and severe hallucinations. This dataset follows strict AAA Quality Guidelines:

• —Semantic Segmentation & Chunking: Raw legal texts and complex case law have been segmented logically by concepts, not arbitrary character limits, ensuring full context retention.
• —Semantic & Prompt Diversity: To prevent over-fitting and model stubbornness, the prompts are engineered across 7 distinct reasoning types (factual lookup, scenario-based edge cases, negative examples, logical reasoning, and legal deduction).
• —Deterministic & Grounded Responses: Every single Assistant output is strictly anchored in the provided context, utilizing structured logical connectors ("Conformément au texte...", "Cependant...", "Par conséquent...") to train the model to maintain a grounded, professional expert persona.
• —Format-Ready: Perfectly formatted .jsonl with explicit System Prompts enforcing strict context-adherence ("Tu es un assistant expert qui répond aux questions en te basant exclusivement sur le contexte fourni...").

📊 Sample Structure

{
  "conversations": [
    { "from": "system", "value": "Tu es un assistant expert..." },
    { "from": "human", "value": "Contexte :\nLe règlement (CE) no 261/2004...\n\nQuestion :\nQuelle est la date d'adoption..." },
    { "from": "gpt", "value": "Conformément au texte, le règlement..." }
  ]
}

💼 Need Bespoke Datasets for Your Domain? (B2B / Freelance)

Are your engineers wasting 70% of their time cleaning data or writing synthetic prompts instead of optimizing architectures and weights? Let me handle the heavy lifting.

I build custom Python extraction workflows, semantic segmentation scripts, and advanced prompt-engineering pipelines to turn your raw data (Internal Docs, Legal PDFs, Financial Reports, Technical Manuals) into clean, production-grade training data.

🔗 Let's Collaborate safely:

• —Secure Project Management & Billing: I use my Fiverr Pro Portal as a secure escrow and compliance platform for corporate contracts.
• —Order Custom Datasets Instantly: Audit my execution and packages directly on my live portal: Get Your Custom AAA Dataset Here
• —Confidentiality: Zero leakage. All proprietary documents are processed inside air-gapped pipelines and destroyed post-delivery.

Disclaimer: This is a public open-source sample dataset for evaluation purposes. All corporate data engineering tasks are executed under strict NDA agreements.