moa-telehealth 1.0.1

Magnetic Outlier Agent: Patent-pending outlier detection for RAG systems and telehealth compliance

Magnetic Outlier Agent (MOA)

Patent Pending • U.S. Provisional Serial No. 63/926,578 filed November 27, 2025
License • Business Source License 1.1 (Change Date: November 27, 2029)
Trademarks Pending • "Magnetic Outlier Agent" and "MOA"

---

Overview

MOA is a geometric multi-metric outlier detection system for embedding vectors in Retrieval-Augmented Generation (RAG) pipelines. It uses a patented triple-metric approach combining centroid distance, local density, and graph curvature to identify low-quality or non-compliant documents before they reach your LLM.

Use cases:

• RAG quality filtering (remove hallucination-prone chunks)
• Telehealth compliance auditing (detect missing consent/risks language)
• Regulatory integration via MCP (ICD-10, CPT billing, state bills)

---

Quick Start

pip install moa-telehealth

Basic Usage

from moa.filters import MOAOutlierFilter

# Initialize filter
moa = MOAOutlierFilter(threshold=1.5)

# Score documents
embeddings = [[0.1, 0.2, ...], ...]  # Your doc embeddings
scores, outliers = moa.score(embeddings)

# Filter out outliers
clean_docs = [doc for i, doc in enumerate(docs) if i not in outliers]

LangChain Integration

from langchain.chains import RetrievalQA
from moa.integrations.langchain_moa_filter import MOAFilterRunnable

qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=vectorstore.as_retriever() | MOAFilterRunnable(MOAOutlierFilter())
)

---

IP & Security Status

• Patent: U.S. Provisional 63/926,578 (filed Nov 27, 2025) – See PATENT_PENDING.md
• License: BUSL 1.1 (change date: 2029-11-27) – See LICENSE
• PyPI: moa-telehealth 1.0.0 live
• MCP Layer 3: Regulatory integration complete – See docs/REGULATORY_MCP_INTEGRATION.md
• Security: PHI handling & key management policies – See SECURITY.md
• Trademarks: Filing in progress (deadline: Dec 4, 2025)

---

Documentation

• Regulatory MCP Integration - Free gov/insurance API integration
• Deployment Summary - Production deployment guide
• Patent Pending - Algorithm claims and scope
• Security - Vulnerability reporting and responsible use

---

Features

Core Algorithm (Patent Pending)

• Triple-metric scorer: Combines centroid distance (d), local density (δ), and graph curvature (κ)
• Bridge-path detection: Identifies weak links in k-NN graphs
• Spectral degradation: Fourier analysis for time-ordered corpora

Integrations

• ✅ LangChain - Drop-in RAG filter
• ✅ MCP - Claude Desktop tools for regulatory lookups
• ✅ FastAPI - RESTful endpoints for web/n8n
• ✅ Streamlit - Interactive compliance editor

Regulatory Layer (MCP)

• ICD-10 lookup (NLM Clinical Tables) - Free, 1000/day
• CPT billing (eCFR §410.78) - Free, unlimited
• State bills (OpenStates) - Free, 5000/day (API key required)
• PHI guard - Automatic blocking of SSN/phone/names

---

License

Business Source License 1.1
© 2025 Michael Ordon. Patent Pending U.S. Provisional 63/926,578. All Rights Reserved.

• Permitted: Evaluation, research, internal use (≤100 visits/month)
• Restricted: Commercial SaaS offerings before change date (2029-11-27)
• Change License: Apache 2.0 (after Nov 27, 2029)

For commercial licensing: [e-mail]

---

Citation

If you use MOA in research, please cite:

@misc{ordon2025moa,
  title={Magnetic Outlier Agent: Geometric Multi-Metric Outlier Detection for RAG Systems},
  author={Ordon, Michael},
  year={2025},
  note={U.S. Provisional Patent Application Serial No. 63/926,578}
}

---

Contact

Inventor: Michael Ordon (Culver City, CA)
Email: [e-mail]
Patent: U.S. Provisional 63/926,578
License: BUSL 1.1
Links: PyPI | Patent Center