issdandavis/phdm-21d-embedding

PHDM 21D Embedding Model

Custom embedding model for the SCBE-AETHERMOORE framework. Maps text inputs into a 21-dimensional Poincare Ball manifold for hyperbolic AI safety governance.

Architecture

• Embedding Dimension: 21D (6D hyperbolic + 6D phase + 3D flux + 6D audit)
• Geometry: Poincare Ball B^n with Harmonic Wall containment
• Polyhedral Lattice: 16 cognitive polyhedra (5 Platonic + 3 Archimedean + 2 Kepler-Poinsot + 2 Toroidal + 4 Johnson/Rhombic)
• Neurotransmitter Weights: Six Sacred Tongues (KO=1.0, AV=1.62, RU=2.62, CA=4.24, UM=6.85, DR=11.09)

Training Data

• Notion knowledge base exports (SCBE technical docs, PHDM specs)
• Perplexity interaction logs (filtered via GeoSeal privacy layer)
• Sacred Tongue tokenized corpora

Deployment

• GCP: Vertex AI Model Registry + GKE Autopilot (test-scbecluser)
• AWS: Lambda functions for intent classification
• HuggingFace: Model weights and inference API

Usage

from phdm_embedding import PHDMEmbedder

embedder = PHDMEmbedder.from_pretrained("issdandavis/phdm-21d-embedding")
vector = embedder.encode("Book a flight from SFO to NYC")
# Returns: 21D numpy array in Poincare Ball coordinates

Dataset Setup (PowerShell)

Use this repo as a working directory for Hugging Face datasets:

cd C:\Users\issda\hf-repos\phdm-21d-embedding
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements-datasets.txt

Set your token in the current shell session:

$env:HF_TOKEN="hf_your_token_here"

Load and preview a dataset split:

python scripts/load_hf_dataset.py --dataset-id issdandavis/scbe-aethermoore-knowledge-base --split train --limit 3

Push local JSONL files to a dataset repo:

python scripts/push_jsonl_dataset.py --dataset-id issdandavis/scbe-aethermoore-knowledge-base --train .\data\train.jsonl --validation .\data\validation.jsonl

Convert Perplexity/Markdown exports into JSONL splits:

python scripts/markdown_to_jsonl.py --input-dir C:\path\to\perplexity-export --output-dir .\data --train-ratio 0.9 --validation-ratio 0.1

One-shot flow (convert then push):

python scripts/markdown_to_jsonl.py --input-dir C:\path\to\perplexity-export --output-dir .\data
python scripts/push_jsonl_dataset.py --dataset-id issdandavis/your-central-knowledge-base --train .\data\train.jsonl --validation .\data\validation.jsonl --test .\data\test.jsonl

Expected JSONL row format example:

{"id":"6e4fcd3f34f5b021","source":"perplexity_space_export","space":"SCBE GitHub Deployment","relative_path":"SCBE GitHub Deployment/notes.md","title":"Deployment Notes","text":"Example source content","meta":{"author":"issdandavis"}}

Related

• SCBE-AETHERMOORE GitHub
• Knowledge Base Dataset
• Interaction Logs Dataset

Model Details

Property	Value
Model Type	Hyperbolic text embedding model
Dimensions	21 (6D hyperbolic + 6D phase + 3D flux + 6D audit)
Geometry	Poincare Ball B^n
Base Model	sentence-transformers/all-MiniLM-L6-v2
Framework	SCBE-AETHERMOORE
License	Apache 2.0

Intended Uses

• AI Safety Governance: Embedding intent vectors for safety classification within the SCBE-AETHERMOORE fleet coordination system
• Semantic Retrieval: Hyperbolic nearest-neighbor search over knowledge base documents
• Intent Classification: Mapping user queries to governance decision nodes
• Anomaly Detection: Identifying out-of-distribution inputs via Poincare Ball boundary proximity

Limitations

• This model is optimized for the SCBE-AETHERMOORE domain and may not generalize to unrelated NLP tasks without fine-tuning
• The 21-dimensional Poincare Ball geometry requires specialized distance metrics (hyperbolic distance, not cosine similarity)
• Currently English-only

Ethical Considerations

This model is designed for AI safety governance. The embedding space encodes hierarchical safety constraints using hyperbolic geometry, ensuring that safety-critical decisions maintain proper containment boundaries. The Harmonic Wall containment mechanism prevents unsafe embeddings from exceeding the Poincare Ball boundary.

Citation

If you use this model, please cite:

@misc{davis2025phdm,
  title={PHDM 21D Embedding: Hyperbolic Embeddings for AI Safety Governance},
  author={Davis, Issac},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/issdandavis/phdm-21d-embedding}
}