Experimentlab/README

⚛️ Experimentlab (Experimentlab.in)

Sovereign Artificial Intelligence for Science, Technology, Engineering & Mathematics (STEM)

"Here’s to the rebels who build real things in the physical world. The ones who refuse to rent their intelligence by the token. Who pull the plug and keep calculating."

---

🌌 The Mission: Sovereign STEM Intelligence

Experimentlab is a deep-tech research company building sovereign foundation models, efficient Small Language Models (SLMs), and air-gapped desktop workstations engineered specifically for physical engineers, roboticists, semiconductor designers, and mathematicians.

Led by Founder & Chief Architect Shreyansh Singh, our research bridges fundamental academic theory with sovereign edge execution:

• Zero Cloud Dependence: 100% offline, privacy-first execution on local laptops and workstations.
• Deterministic Mathematical Truth: Computer Algebra System (SymPy CAS) verification eliminating hallucinated calculus, ODEs, and physics.
• Sovereign Quantization: Pioneering Sparse Mixture of Experts (MoE) and 1.58-bit ternary {-1, 0, +1} BitNet quantization running natively on edge CPUs and NPUs.

---

🏛️ The 5-Pillar Public AI Catalog

graph LR
    A["Raw STEM Corpora (1B+ Tokens, arXiv, Dolma)"] --> B["Vigyan Sovereign Pretraining"]
    B --> C["1. Frontier STEM SLMs (32B / 7B / 2B)"]
    B --> D["2. Advanced Architectures (MoE / 1.58-bit / GRPO)"]
    B --> E["3. Domain Mastery (RTL / Math / Biomed / Vidhi)"]
    C --> F["5. Edge GGUF Runtimes & Desktop Studios"]
    D --> F
    E --> F
    B --> G["4. Foundation Pretraining Datasets"]

1. 🚀 Frontier STEM & Dense SLMs

• Vigyan-AI-32B & Vigyan-AI-32B-Titan-v1: Flagship 32B Titan models for deep multi-step scientific deduction.
• Vigyan-7B-STEM-Instruct-v1 & Shreyansh-STEM-AI-7B-Final: High-performance 7B dense scientific models.
• Shreyansh-STEM-AI-2B-v3 (1,195+ downloads) & Shreyansh-STEM-AI-2B-v3-Distilled: Compact, battery-friendly edge SLMs.

2. 🌟 Advanced Architectures (MoE, 1.58-Bit Ternary, GRPO)

• Vigyan-1.5B-4x-MoE (1,669+ downloads): 4-expert routed Sparse MoE delivering 7B-class STEM reasoning at 1.5B inference latency.
• Vigyan-3B-4x-MoE (704+ downloads): Advanced Top-1 routing multi-expert model tuned for symbolic logic.
• vigyan-3b-1.58bit-qat: Pure 1.58-bit {-1, 0, +1} ternary BitNet model trained via Quantization-Aware Training (QAT). Replaces matrix multiplication with integer additions.
• vigyan-2b-reasoning-grpo: DeepSeek-style GRPO test-time reasoning SLM trained on mathematical and logical verification trees.
• vigyan-olmoe-1b-7b-masterpiece-adapter: Ai2 OLMoE sparse expert architecture.

3. 🔬 Doctoral & Domain-Specialized Intelligence

• olmo2-7b-silicon-rtl-eda: Hardware silicon EDA specialist for synthesizable IEEE 1364 Verilog RTL, FSMs, and synchronous timing circuits.
• olmo2-7b-phd-pure-math: Specialization in pure mathematics, formal topological proofs, and abstract algebra.
• olmo2-7b-biomed-chem: Deep biochemistry, enzyme kinetics, and molecular thermodynamics.
• olmo2-7b-astro-logic: Orbital mechanics, Hohmann transfers, and astrophysical physics.
• Vidhi-AI-Instruct: Sovereign Indian legal intelligence aligned with the Bharatiya Nyaya Sanhita (BNS).

4. 📚 Foundational Pretraining Corpora & Datasets

• shreyansh-1B-SLM-pretrain-stem-english: 1 Billion token foundation pretraining corpus balanced for cold-start STEM intelligence.
• shreyansh-hinglish-english-stem-500k: 500,000 sample bilingual Hinglish-English dataset tailored for Indian engineering students.
• vigyan-ai-research-papers: Curated corpus of scientific literature and verified technical preprints.

5. ⚡ Edge GGUF Runtimes & Workstation Integration

• Vigyan-Models-GGUF: Quantized GGUF binaries (Q4_K_M) engineered for local execution via llama.cpp and Ollama.
• vigyan-7b-moe-gguf: Edge quantized MoE binary.
• Vigyan Workstation: Open-source desktop engineering studio pairing these weights with 3D robotics kinematics, WaveDrom waveforms, and KiCad PCB netlists.

---

⚖️ Intellectual Property & Strict Non-Commercial Licensing

All open-source models, benchmark suites, pretraining datasets, research whitepapers, and software released across the Vigyan / ExperimentLab ecosystem are strictly governed by the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license and the ExperimentLab Sovereign Community & Demilitarized License:

• Academic, Educational & Personal Research Use Only: Freely permitted for students, academic researchers, university coursework, self-hosting, and civilian scientific inquiry.
• Strictly Non-Commercial Prohibition: You may not sell, monetize, host as a paid API/SaaS wrapper, bundle with commercial hardware, or use for commercial advantage without prior written enterprise authorization from ExperimentLab AI.
• Demilitarized Civilian STEM Restriction: Strictly prohibited for military, defense, weapons research, autonomous combat, or surveillance systems.
• Commercial Enterprise Licensing: For commercial licensing, enterprise deployment, or institutional partnerships, contact contact@experimentlab.in.

---

📜 Academic Attribution & Sovereign Citation

@article{singh2026vigyan,
  title={Vigyan Sovereign Trilogy: Empirical Scaling Laws and Deterministic Symbolic-AST Evaluation of Edge Small Language Models},
  author={Singh, Shreyansh},
  institution={Vigyan AI Open Research},
  journal={Vigyan AI Research Publications},
  year={2026},
  url={https://github.com/shreyansh001boy-tech/vigyan-sovereign-benchmark}
}

---

_{Crafted with passion for sovereign open-source STEM research by ExperimentLab.in • Built for the Crazy Ones.}