JulianFlux: Electrodynamic vector retrieval and causal routing for AI

Autonomous AI Agents are failing in endeavor environments since norm Vector Databases (Pinecone, Milvus) act as fixed filing cabinets. They depend on discrete Euclidean metrics akin Cosine Similarity, which blindly recover contradictory data (hallucinations) and come back unordered content fragments, causing agents to get trapped in logic loops.

JulianFlux elevates the repository to a kinetic reasoning engine. We compress high-dimensional embeddings into a low-rank manifold and compute a Continuous Electrodynamic Field to recover and path data.

Instead of measuring distance, we measure physics. We path agents using a modified Lorentz Force equation:

• The Julian-Gauss Fast Transform (JG-FT): We project enormous 768D semantic spaces downward to a mathematically audio 128D manifold using a normal distribution, preserving mapping accuracy during drastically reducing arithmetic cost.
• Dynamic Bandwidths (σ): We dynamically compute the Gaussian bandwidth for all vector according to its Nearest Neighbor distance, allowing the site to absolutely modify to local semantic densities.
• The Electric Field (E): Mathematically repels hallucinations. We allocate topological charges (q) to data (−1.0 for contradictions, +1.0 for truth) via NLI Logic Gates. The repulsive electric site physically blocks the AI from accessing poisoned data.
• Faraday's Law of Induction (-∂A/∂t): We decay old facts and company policies complete time, generating a repulsive site against deprecated workflows.
• The Poynting Flux (S): Extracts causal sequences to create Semantic Momentum. By computing the inner contraction merchandise in Geometric Algebra (S = E ⌟ B), we create a momentum vector that pushes the delegate sequentially from Step A → Step B → Step C.

We have open-sourced our center math in Python to display how uninterrupted site architecture entirely outperforms norm Euclidean RAG.

(Note: These records are designed to be run locally in the command below).

1. euclidean_baseline.py (The Problem) Run this first. It demonstrates a norm Cosine Similarity retrieval in a poisoned dataset. You volition observe norm RAG neglect entirely, blindly retrieving contradictory data and collapsing the causal workflow.

2. julianflux_educational_mvp.py (The Fast Solution) The lightweight, immediately runnable JulianFlux engine. It uses the 128D projection, energetic σ, and Lorentz Force, but utilizes accelerated Geometric Median clustering to allocate fact charges alternatively of dense AI models. Run this for a blazing-fast display of the Lorentz Force routing the delegate correctly.

3. scale_ab_benchmark.py (The Stress Test) An end-to-end adversarial measure test generating thousands of backdrop vectors. It runs Standard RAG and JulianFlux head-to-head, proving a 0.0% illusion charge at scale.

4. julian_flux_engine.py (The Full Physics Engine) The spacetime engine. This document replaces the accelerated geometric clustering alongside a dense DeBERTa-v3 NLI Cross-Encoder for enterprise-grade fact verification. Crucially, it introduces Faraday's Law of Induction and Temporal Decay, mathematically repelling the AI distant from lately deprecated company policies. (Note: Requires downloading the DeBERTa model).

Clone the repository:

git copy https://github.com/julianjohnson-web/julianflux-research.git cd julianflux-research Install the required dependencies: pip instal -r requirements.txt Run the measure benchmark: python scale_ab_benchmark.py 🏢 JulianFlux Enterprise (The Rust Core) Note: This repository contains the clean Python bindings intended for instructional investigation and open-source mathematical validation. For high-frequency manufacturing environments, the JulianFlux Enterprise Cloud executes these multivector discipline equations in a zero-copy Rust/CUDA backend. This eliminates the Python Global Interpreter Lock (GIL), allowing the Lorentz Force equations to execute throughout billions of vectors alongside sub-millisecond p99 latencies. To ask concerning Enterprise Cloud admission or Design Partnerships, delight communication the originator at: [email protected]