/ The Cognitive Loop

Three cognitive stages, one query.

01 — Interpret

Intent Understanding

Rexxengine decodes what you actually mean — parsing entities, constraints and the shape of the answer you need before a single source is touched.

02 — Retrieve

Semantic Retrieval

A hybrid engine fuses lexical precision with dense semantic recall, surfacing sources by meaning and explanatory depth — never keyword noise.

03 — Connect

Cognitive Link Generation

Each result is rebuilt as a cognitive link: summarised, tagged, scored for reliability and wired into a live knowledge graph of relations.

/ How Rexxengine Works

A manifesto in four movements.

01

Parse the intent

Rexxengine reads your query the way a researcher reads a brief — extracting entities, intent type and the answer's expected shape. A link and a question are treated as the same thing: a request to understand.

02

Retrieve by meaning

A hybrid retriever runs lexical and semantic passes in parallel across the corpus, ranking candidates by explanatory depth and conceptual proximity rather than surface term frequency.

03

Reason over the graph

Retrieved evidence is projected into a knowledge graph. Rexxengine traces relations between concepts, resolves contradictions and clusters sources into coherent lines of reasoning.

04

Return cognitive links

Every answer is composed as structured cognitive links — summary, key concepts, reliability score, relation graph and source — so you leave with understanding, not a list of tabs.

/ Cognitive Results

Results that think back.

Recognized Intent

Comparative Research

confidence 94%

Entities

TransformersSelf-AttentionContext WindowsEmbeddings

Attention Is All You Need — The Architectural Foundation

CL-01

The seminal work replacing recurrence with pure self-attention. Every token attends to every other token in parallel, weighted by learned query-key similarity. Multi-head attention captures different relational subspaces at once, and the result is a model that understands context through weighted relationships rather than sequential memory.

Key concepts

Self-AttentionMulti-HeadPositional Encoding

Reliability

98%

Relations

↳ arxiv.org / 1706.03762

/ Knowledge Graph

Every answer is a living map.

Rexxengine doesn't return a list — it returns a structure. Concepts become nodes, relations become edges, and reasoning becomes something you can see and navigate.

  • Neon-cyan concept nodes with live glow on activation
  • Electric-blue edges trace reasoning between concepts
  • Hover any node to illuminate its relational neighbourhood
Query IntentSemanticsLexicalKnowledge GraphRankingEmbeddingsRelationsSourcesClusters
● active node — Query Intent

/ Technology & Architecture

An engine built to reason.

Rexxengine architecture

Rexxpons Cognitive Core

Retrieval → Reasoning → Ranking

BM25Dense VectorsGraph DBRerankerLLM Layer

Hybrid Retrieval

Lexical BM25 precision fused with dense vector recall — two passes, one ranked candidate set.

Cognitive Intent Analysis

A reasoning layer classifies intent, extracts entities and predicts the answer's ideal structure.

Knowledge Graph Reasoning

Evidence is projected into a graph where relations are traced and contradictions resolved.

Cognitive Ranking Model

Sources are scored on explanatory depth and reliability, not keyword frequency.

/ Brand Family

Powered by the Rexxpons Cognitive Framework

Rexxengine is one instrument in the Rexxpons family — a shared foundation for machines that understand.

Cognitive Retrieval
Semantic Reasoning
Knowledge Graphs
Intent Analysis
Structured Answers
Cognitive Retrieval
Semantic Reasoning
Knowledge Graphs
Intent Analysis
Structured Answers