EngineeringWays: Circuit Analysis Reasoning Dataset (LoRA Ready)

Skip the scraping, cleaning, and formatting. Start fine-tuning immediately.

The EngineeringWays Circuit Analysis Reasoning Dataset is a commercially viable, strictly deduplicated, and rigorously formatted JSONL dataset designed specifically for training Large Language Models (LLMs) in advanced STEM reasoning.

Built for fine-tuning models like Llama 3, Mistral, and DeepSeek, this dataset forces the AI to learn the underlying algebra and physics of electrical engineering rather than just memorizing answers.

Standard License Scope

This specific tier is designed for indie developers, academic researchers, and small startups looking to build and train models affordably.

• Commercial & Training Rights: Cleared for internal research, educational purposes, and training machine learning models. You are legally permitted to monetize the models you train, provided you meet the strict eligibility limits below.

• Eligibility Limits: To qualify for this tier, you or your organization must have fewer than ten (10) employees and generate less than $1,000,000 USD in annual gross revenue. (If you exceed these limits, you must purchase an Enterprise tier).

• Data Restrictions: As with our standard terms, you cannot resell the raw dataset, upload it to public repositories (like GitHub or Hugging Face), or use it to generate derivative/synthetic datasets to sell.

 View the full Master Dataset License Agreement here: [Click here]

What makes this dataset premium?

• Chain-of-Thought (<think> tags): Every problem includes an internal scratchpad/monologue where the AI breaks down the KVL/KCL equations, identifies supernodes/supermeshes, and self-corrects before outputting the final answer.
• Strict Textbook Formatting: All math and final answers are perfectly wrapped in industry-standard LaTeX ($...$ and $$...$$ ).
• Zero Hallucinations: Manually verified for algebraic and physical accuracy.
• Aggressively Deduplicated: Filtered down from raw generation to 592 perfectly unique, high-quality items. No exact copy-pastes, no low-quality or suspiciously short outputs.

Technical Specs:

• Format: .jsonl (JSON Lines) - Ready for immediate Hugging Face / LoRA pipeline ingestion.
• Row Count: 592 Unique Items
• Domain: Electrical Engineering, Circuit Analysis, Nodal/Mesh Analysis, Op-Amps, Thevenin Equivalents.
• Structure: Standard {"instruction": "...", "input": "...", "output": "<think>...</think>..."} format.

Try before you buy: Download a free 50-item sample on our Kaggle/Hugging Face/GitHub page!