neuroembed 0.1.1

Model-agnostic semantic embedding enrichment framework

🧠 NeuroEmbed

NeuroEmbed is a model-agnostic semantic embedding enrichment framework.

It does not replace embedding models.
Instead, it modulates embeddings using semantic context, producing controlled directional shifts in vector space while preserving dimensionality and normalization.

Designed for:

• RAG systems
• Conversational memory
• Knowledge-aware retrieval
• Agent architectures
• Local / offline-first AI systems

---

⭐ Why NeuroEmbed Exists

NeuroEmbed was built to explore context-aware vector representations in a simple, testable, and model-independent way — without hype or overengineering. If you work on RAG, memory systems, or agent architectures, NeuroEmbed is designed to be a clean building block you can trust.

❌ What NeuroEmbed Is NOT

• Not a vector database
• Not a retriever
• Not a model replacement
• Not a “state-of-the-art accuracy” claim

NeuroEmbed is a semantic modulation layer, designed to integrate cleanly into existing systems.

🏗️ Architecture Overview

Text Input
   │
   ▼
[ Base Encoder ]
   │
   ▼
Base Embedding  ──────────────┐
                              │
Context Texts ─▶ Encoder ─▶ Context Mean
                              │
                              ▼
                  Context Injector (α)
                              │
                              ▼
                   Enriched Embedding

GitHub: https://github.com/Umeshkumar667/NeuroEmbed

🚀 Installation

Standard install (recommended)

pip install neuroembed

⚡ Quick Start

from neuroembed.core import NeuroEmbed
from neuroembed.encoders.sentence_transformer import SentenceTransformerEncoder

# Initialize encoder (replaceable)
encoder = SentenceTransformerEncoder()

# Initialize NeuroEmbed
ne = NeuroEmbed(encoder=encoder, alpha=0.6)

# Input text
query = "bank interest rate"

# Optional semantic context
context = [
    "RBI monetary policy",
    "repo rate",
    "inflation control"
]

# Generate enriched embedding
embedding = ne.embed(query, context)

print("Embedding shape:", embedding.shape)
print("Embedding norm:", (embedding @ embedding) ** 0.5)

embedding = ne.embed("hello world")

base = encoder.encode(["bank interest rate"])[0]
enriched = ne.embed("bank interest rate", context)

print("Cosine similarity:", base @ enriched)