tantra-ai-labs/high-reasoning-dataset-v1

High-Reasoning Dataset v1

2,139 premium Q&A pairs autonomously generated by a multi-AI knowledge distillation system. Each answer includes deep reasoning (CoT), mathematical foundations, production-ready code, historical context, edge cases, and real-world production incidents.

Made by [Aditya Wakharkar](https://github.com/codewith-aditya) | Tantra AI Labs

What is this dataset?

This is a synthetic training dataset created entirely by two AI models talking to each other 24/7 — no humans in the loop. One model acts as a relentless PhD researcher (asking questions, never answering), the other as the world's best teacher (giving comprehensive, multi-section answers). The result is high-quality reasoning data suitable for fine-tuning LLMs.

Who made this?

[Aditya Wakharkar](https://github.com/codewith-aditya) — AI Engineer & Founder at Tantra AI Labs. I design and build autonomous AI systems that push the boundaries of what's possible with synthetic data generation. This dataset is the first public release from the Tantra AI Labs knowledge distillation platform.

• —GitHub: github.com/codewith-aditya
• —LinkedIn: linkedin.com/in/aditya-wakharkar-29ab10321

How the Dataset Was Created

Two frontier AI models run in an infinite academic loop:

Every sample passes through a 7-dimension quality verifier before being written to disk with immediate sync (crash-safe).

What Makes This Dataset Special

Unlike most synthetic datasets:

• —✅ Every answer has reasoning + math + code + edge cases — not just text
• —✅ Zero repeated questions — PhD agent tracks full history
• —✅ 12-section structure per answer enforced by prompt
• —✅ Quality verified across 7 dimensions before storage
• —✅ Crash-safe 24/7 operation — dataset grows continuously

What Can You Use This For?

Dataset Structure

Dataset Stats

Topic Distribution

Answer Quality — 12 Required Sections

Every single answer includes all of these:

1. 1.Thinking & Approach — mental models, frameworks, step-by-step plan
2. 2.Historical Background — origins, evolution, key contributors, milestones
3. 3.Detailed Explanation — first principles, terminology, core concepts
4. 4.Mathematical Foundations — full derivations, proofs, worked numerical examples
5. 5.Implementation & Code — production-ready Python with error handling
6. 6.Production Architecture — deployment, scaling, monitoring, disaster recovery
7. 7.Edge Cases & Failures — what breaks, debugging strategies, recovery
8. 8.Optimization Strategies — performance, cost, profiling, tuning
9. 9.Alternatives & Comparisons — industry alternatives, trade-offs, migration
10. 10.Common Misconceptions — interview traps, what people get wrong
11. 11.Research & Further Reading — papers, books, open problems
12. 12.Production Incidents — real failures, root cause analysis, lessons learned

Generation Architecture

Seed Topics → Topic Discovery → Knowledge Graph Coverage Check
                                       ↓
                              PhD Research Agent (6 depth levels)
                              ├── Anti-repetition (full question history)
                              ├── Critical analysis of every answer
                              └── Challenge generation for weak spots
                                       ↓
                              Master Teacher (GLM-5)
                              └── 12-section comprehensive answers
                                       ↓
                              7-Dimension Quality Verification
                              ├── Completeness
                              ├── Correctness
                              ├── Reasoning Depth
                              ├── Implementation Depth
                              ├── Novelty
                              ├── Factual Consistency
                              └── Hallucination Probability
                                       ↓
                              Raw TXT + JSONL Storage (crash-safe)
                                       ↓
                              Hugging Face Dataset

The entire system runs on a single VPS with:

• —Redis-backed distributed task queues
• —Knowledge Graph with JSON persistence for coverage tracking
• —Periodic checkpointing for crash recovery
• —PM2 process management for 24/7 uptime

Limitations & Future Improvements

Quick Start

from datasets import load_dataset

dataset = load_dataset("adityawakharkar/high-reasoning-dataset-v1")

# Explore
for sample in dataset["train"].take(3):
    print(f"Topic: {sample['topic']}")
    print(f"Question: {sample['question'][:200]}...")
    print(f"Answer length: {len(sample['answer'])} chars")
    print("---")

Citation

@dataset{high-reasoning-v1,
  author = {Aditya Wakharkar},
  title = {High-Reasoning Dataset v1},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/adityawakharkar/high-reasoning-dataset-v1}},
  note = {Autonomously generated by multi-AI knowledge distillation system. Tantra AI Labs.},
}

License

MIT — use freely for research and commercial purposes.

Built with relentless AI agents by [Aditya Wakharkar](https://github.com/codewith-aditya) | [Tantra AI Labs](https://github.com/tantrailabs)