Hanzo AI Research Papers

Academic and technical papers for Hanzo AI compute infrastructure and protocols.

Organization: https://github.com/hanzoai Website: https://hanzo.ai Contact: contact@hanzo.ai

---

📚 Overview

Hanzo AI research is organized into 5 comprehensive papers covering the complete AI compute infrastructure stack (Layer-1) that enables decentralized model training and inference.

---

📄 Papers Collection (5 Papers)

1. Hanzo ASO (Active Semantic Optimization)

File: hanzo-aso.tex → hanzo-aso.pdf HIP: HIP-002-aso.md Status: ✅ Published October 2025

Title: "Training-Free Adaptation via Active Semantic Optimization and Product-of-Experts Decoding"

Abstract: A training-free adaptation framework for agentic code generation built on TF-GRPO and PoE decoding.

Key Contributions:

• Training-Free GRPO (TF-GRPO) with epistemic utility
• Product-of-Experts (PoE) decoding at token level
• 1-bit semantic compression (BitDelta) - 29.5× savings
• Hanzo Dev CLI agent with SWE-bench integration
• 18.2% resolved rate on SWE-bench Verified

Sections: tf-grpo.tex, poe-decoding.tex, bitdelta.tex, swe-bench-eval.tex

---

2. Hanzo DSO (Decentralized Semantic Optimization)

File: hanzo-dso.tex → hanzo-dso.pdf HIP: HIP-003-dso.md Status: ✅ Published October 2025

Title: "Decentralized Semantic Optimization with Byzantine-Robust Prior Aggregation"

Abstract: A protocol for sharing and aggregating experiential priors across distributed language model agents without parameter updates.

Key Contributions:

• Byzantine-robust median voting with stake weighting
• ExperienceRegistry smart contract (IPFS/Arweave storage)
• P2P gossip protocol for prior synchronization
• Quality scoring and slashing mechanism
• 15.2% improvement in multi-agent tasks vs isolated operation

Sections: dso-core.tex, bitdelta.tex

---

3. Hanzo HMM (Hamiltonian Market Maker)

File: hanzo-hmm.tex → hanzo-hmm.pdf HIP: HIP-004-hmm.md Status: ✅ Published October 2025

Title: "Hamiltonian Market Maker for Decentralized AI Compute Exchange"

Abstract: An automated market maker for pricing heterogeneous AI compute resources via conserved Hamiltonian invariants.

Key Contributions:

• Hamiltonian invariant H(Ψ,Θ) = κ for oracle-free pricing
• Multi-asset routing with SLA-aware path solver
• Risk-adjusted fee structure for inventory management
• PoAI integration for verifiable job settlement
• < 200ms quote latency, 98.7% price stability (vs 89.2% oracle-based)

Sections: hmm.tex, poai.tex, token-economics.tex

---

4. Hanzo Network Architecture

File: hanzo-network-architecture.tex → hanzo-network-architecture.pdf Status: ✅ Published

Title: "Hanzo Network: Decentralized AI Compute Infrastructure"

Abstract: Complete architectural specification of Hanzo's Layer-1 compute infrastructure, including consensus mechanism, TEE attestation, GPU node management, and integration with Lux (L0) and Zoo (L2).

Key Contributions:

• Layered architecture design (Lux → Hanzo → Zoo)
• Self-mining consensus (0 token requirement for validators)
• GPU compute verification via Lux A-Chain TEE attestation
• Integration with Zoo's Experience Ledger and HLLM framework
• Multi-GPU support (tensor/pipeline/sequence parallelism)

---

5. Hanzo Network Whitepaper

File: hanzo-network-whitepaper.tex → hanzo-network-whitepaper.pdf Status: ✅ Published

Title: "Hanzo Network: Economic Model and Tokenomics"

Abstract: Comprehensive overview of Hanzo's economic model, validator incentives, and governance mechanisms.

Key Contributions:

• Self-mining model (validators earn through compute contribution)
• Integration with HMM for dynamic pricing
• Cross-chain economic settlement via Lux Bridge
• Governance framework for network parameters

---

🌐 Cross-Ecosystem Research

The Lux-Hanzo-Zoo-Zen ecosystem has published 58 comprehensive research papers:

• Lux (L0): 24 papers on consensus, post-quantum crypto, DeFi, cross-chain
• Hanzo (L1): 5 papers on compute infrastructure, ASO/DSO, HMM
• Zoo (L2): 7 papers on AI training, tokenomics, HLLM
• Zen: 22 papers on efficient LLMs with spatial reasoning

Cross-Layer Innovations

Validator Economics:

• Lux validators: 1M LUX stake (PoS + Genesis, high security, L0 foundation)
• Hanzo validators: 1 AI token (PoW compute, self-mined on any device, participate in HMM market)
• Zoo validators: 1,000 ZOO stake (PoAI - weighted by LLM experience sharing and semantic contributions)

Research Integration:

• Lux A-Chain provides TEE attestation for Hanzo compute verification
• Hanzo ASO/DSO powers Zoo's Training-Free GRPO implementation
• Hanzo HMM enables economic settlement for Zoo Experience Ledger
• Zen models (7680-dim embeddings) serve as base frontier models

---

🔗 Related Papers by Ecosystem

Lux Network (Base Layer - L0)

24 foundational papers covering consensus, post-quantum cryptography, DeFi, and cross-chain:

• Consensus: Multi-consensus, Quantum, Quasar, FPC (4 papers)
• Chain Architecture: A-Chain (TEE), G-Chain (GraphQL), M-Chain (MPC), Z-Chain (Privacy) (4 papers)
• DeFi: Lightspeed DEX, Credit Lending, Oracle, Perpetuals (4 papers)
• Web3: NFT Market, ID IAM, DID Specification (3 papers)
• Governance: DAO frameworks (2 papers)
• Post-Quantum: NTT, ETHFALCON, Threshold Signatures (3 papers)
• Scalability: Verkle Trees, Fraud Proofs, TEE Mesh (3 papers)

Zoo Network (AI/ML Specialization - L2)

7 papers on AI training infrastructure and tokenomics:

• Foundation Mission: 501(c)(3) non-profit charter
• Network Architecture: Layered AI infrastructure
• Tokenomics: $AI token economics & governance
• Gym Platform: AI training infrastructure
• HLLM Training-Free GRPO: Hamiltonian LLM framework
• Experience Ledger DSO: Semantic memory system
• ZIP-002 Zen Reranker: Cross-model ranking

Zen Language Models (Base Frontier Models)

22 papers on efficient LLMs with spatial reasoning:

• Family Overview: Complete ecosystem (600M-480B params)
• Core Models (6): Nano, Eco, Coder, Omni, Next, Guard
• Creative Models (4): Artist, Artist-Edit, Designer-Instruct, Designer-Thinking
• Specialized Models (7): Scribe, Director, Foley, Musician, Video, Voyager, World
• Advanced (4): 3D, Agent, Technical, Reranker
• Complete listing: Zen Papers Repository

---

🛠️ Building Papers

Requirements

• LaTeX distribution (TeX Live, MacTeX, BasicTeX) OR
• Docker/Colima for containerized builds

Quick Build

# Using Docker (auto-detected if LaTeX not installed)
cd papers
make

# View PDF (macOS)
make view

# Clean intermediate files
make clean

Native LaTeX (Recommended for Regular Use)

# Install LaTeX (macOS)
brew install --cask basictex
# or
make install-latex

# Add to PATH
echo 'export PATH="/Library/TeX/texbin:$PATH"' >> ~/.zshrc
source ~/.zshrc

# Install additional packages
sudo tlmgr update --self
sudo tlmgr install collection-latexextra

# Build
make

Docker Build

# Start Docker daemon (if using Colima)
colima start

# Build (Docker auto-detected)
make

---

📦 Makefile Commands

# Build all papers
make                  # Build all PDFs (ASO, DSO, HMM, Architecture, Whitepaper)
make all              # Same as above

# Build individual papers
make aso              # Build hanzo-aso.pdf only
make dso              # Build hanzo-dso.pdf only
make hmm              # Build hanzo-hmm.pdf only
make architecture     # Build hanzo-network-architecture.pdf only
make whitepaper       # Build hanzo-network-whitepaper.pdf only

# Cleaning
make clean            # Remove intermediate files (.aux, .log, etc.)
make distclean        # Remove all generated files including PDFs

# Utilities
make view             # Open all PDFs (macOS)
make help             # Show all targets with descriptions
make docker-pull      # Pull Docker image (first time setup)

---

📝 LaTeX Packages Used

All papers use standard LaTeX packages included in TeX Live:

• Math: amsmath, amssymb, amsthm, mathtools, bm
• Graphics: graphicx, xcolor
• Tables: booktabs, multirow
• Algorithms: algorithm, algpseudocode
• Navigation: hyperref (with colored links)
• Lists: enumitem
• Layout: geometry

---

🤝 Contributing

Adding New Papers

1. Create LaTeX source file paper-name.tex
2. Update Makefile if needed
3. Build PDF: make
4. Update this README with paper description
5. Commit both .tex source and .pdf output

Paper Style Guidelines

• Use 11pt article class
• 1-inch margins (geometry package)
• Colored hyperlinks (black text, blue citations/URLs)
• Algorithms in pseudocode format
• Tables with booktabs styling
• Include abstract and conclusion
• Provide Solidity interfaces for protocol papers
• Include implementation plan/roadmap

---

📂 File Organization

papers/
├── README.md                        # This file
├── Makefile                         # Build automation (multi-paper support)
├── .gitignore                       # LaTeX artifacts
│
├── hanzo-aso.tex                    # ASO paper source
├── hanzo-aso.pdf                    # ASO paper PDF (7 pages)
├── hanzo-dso.tex                    # DSO paper source
├── hanzo-dso.pdf                    # DSO paper PDF (6 pages)
├── hanzo-hmm.tex                    # HMM paper source
├── hanzo-hmm.pdf                    # HMM paper PDF (7 pages)
├── hanzo-network-architecture.tex   # Architecture paper source
├── hanzo-network-architecture.pdf   # Architecture paper PDF
├── hanzo-network-whitepaper.tex     # Whitepaper source
├── hanzo-network-whitepaper.pdf     # Whitepaper PDF
│
└── sections/                        # Reusable LaTeX sections (shared across papers)
    ├── tf-grpo.tex                  # Training-Free GRPO formulation
    ├── poe-decoding.tex             # Product-of-Experts decoding
    ├── bitdelta.tex                 # 1-bit compression (BitDelta)
    ├── swe-bench-eval.tex           # SWE-bench evaluation protocol
    ├── dso-core.tex                 # DSO protocol specification
    ├── hmm.tex                      # Hamiltonian Market Maker mechanics
    ├── poai.tex                     # Proof of AI attestations
    └── token-economics.tex          # Token economics

Modular Design: Papers use \input{sections/...} to share common sections, reducing duplication and ensuring consistency across the research ecosystem.

---

📖 Citation

If you use Hanzo in your research, please cite the relevant paper(s):

@techreport{hanzo2025aso,
  title={Training-Free Adaptation via Active Semantic Optimization and Product-of-Experts Decoding},
  author={Hanzo Industries Inc.},
  year={2025},
  month={October},
  institution={Hanzo Industries Inc.},
  address={995 Market St, San Francisco, CA},
  note={HIP-002},
  url={https://github.com/hanzoai/papers}
}

@techreport{hanzo2025dso,
  title={Decentralized Semantic Optimization with Byzantine-Robust Prior Aggregation},
  author={Hanzo Industries Inc.},
  year={2025},
  month={October},
  institution={Hanzo Industries Inc.},
  address={995 Market St, San Francisco, CA},
  note={HIP-003},
  url={https://github.com/hanzoai/papers}
}

@techreport{hanzo2025hmm,
  title={Hamiltonian Market Maker for Decentralized AI Compute Exchange},
  author={Hanzo Industries Inc.},
  year={2025},
  month={October},
  institution={Hanzo Industries Inc.},
  address={995 Market St, San Francisco, CA},
  note={HIP-004},
  url={https://github.com/hanzoai/papers}
}

Cross-ecosystem citations:

• Lux: https://github.com/luxfi/papers (24 papers on L0 infrastructure)
• Zoo: https://github.com/zooai/gym/tree/main/papers (7 papers on AI/ML)
• Zen: https://github.com/zenlm/papers (22 papers on efficient LLMs)

---

🔗 Related Projects

Hanzo AI Projects

• Hanzo Node: https://github.com/hanzoai/hanzo (Rust blockchain implementation)
• Agent SDK: https://github.com/hanzoai/agent (Python multi-agent framework)
• MCP Tools: https://github.com/hanzoai/mcp (Model Context Protocol)
• Jin: Multimodal LLM architecture (text/vision/audio/3D)

Zoo Labs Foundation (Partner Organization)

• Zoo Papers: https://github.com/zooai/gym/tree/main/papers
• Zoo ZIPs: Zoo Improvement Proposals for decentralized learning protocols
• DSO (ZIP-001): Decentralized Semantic Optimization - Byzantine-robust prior aggregation
  • Built on Hanzo's ASO (HIP-002) and HMM (HIP-004)
  • Years of co-development between Hanzo AI Inc and Zoo Labs Foundation (501c3)

ZenLM (Co-developed Models)

• Zen Models: https://github.com/zenlm/papers
• Base Frontier Models: Shared foundation for both Hanzo and Zoo ecosystems
• Partnership: Hanzo AI Inc (Techstars '17) & Zoo Labs Foundation (501c3)

---

📜 License

Papers are published under Creative Commons Attribution 4.0 International (CC BY 4.0).

Code examples and implementations referenced in papers follow their respective project licenses.

---

📧 Contact

• Email: contact@hanzo.ai
• Website: https://hanzo.ai
• GitHub: https://github.com/hanzoai
• Twitter: @hanzoai

---

Hanzo Industries Inc. 995 Market St, San Francisco, CA https://hanzo.ai

---

Last Updated: January 28, 2025 Total Papers: 5 Status: Active Development

Building decentralized AI compute infrastructure for the next generation of language models.