Noesis: Bidirectional Graph-RAG with Adaptive Parallelism and Cross-Knowledge-Base Semantic Discovery

Nicola Cogotti

cs.IRAug 16, 2026

Noesis: Bidirectional Graph-RAG with Adaptive Parallelism and Cross-Knowledge-Base Semantic Discovery

Authors: Nicola Cogotti

Organizations: Alpha Cogs

Abstract

Retrieval-Augmented Generation over knowledge graphs (Graph-RAG) has emerged as a powerful paradigm for grounding large language models in domain-specific corpora. However, existing systems face persistent limitations: (1) static chunking fragments long documents, losing cross-section semantic connections; (2) ingestion pipelines do not scale adaptively; and (3) multi-domain deployments require either a monolithic knowledge base that dilutes retrieval precision or manual user routing. We present Noesis, a decoupled Graph-RAG architecture addressing these limitations through four algorithms: (a) Bidirectional Graph Traversal with a Graph-Feedback Context Resolver simulating human reading with degrading memory; (b) an AIMD Concurrency Controller adapted from TCP congestion control, achieving 23x speedup with zero OOM events; (c) Moesis, domain-aware selective quantization for MoE models achieving 6.3x speedup on 12 GB consumer GPUs; and (d) Mesh, cross-KB semantic routing with runtime structural discovery enabling small on-premises models to perform multi-hop cross-domain reasoning. On HotpotQA (1,000 questions), Noesis achieves 59.5 EM / 74.7 F1, surpassing GraphRAG by +27.8 EM while using a 35B on-premises model for graph construction rather than GPT-4o. Source text verification on a 193-page document confirms 90% precision on long-range causal edges inaccessible to chunk-independent extraction.

Figures & tables

Figure 1: Noēsis system architecture. Five decoupled components communicate via REST APIs and Redis message queues, enabling independent scaling and backend replacement.

Figure 2: Bidirectional graph traversal: the forward pass builds the graph with degrading memory context, then the backward pass reconnects early low-degree nodes to semantically related late nodes.

Figure 3: Intra-document cross-linking enabled by bidirectional traversal. The dense interconnection pattern between concepts from different document sections is a direct result of the backward pass reconnection. Single-pass extraction produces sparse, fragmented subgraphs without these long-range edges.

Figure 4: Moēsis three-level pipeline: (1) domain-aware profiling identifies hot/cold experts, (2) selective quantization applies different precision per layer, (3) runtime re-adaptation re-profiles when the domain changes.

Metric	RTX 4080 Laptop (12 GB)	RTX 3060 (6 GB) [stress test]
Prompt processing	$22\to 139$ t/s ( $6.3\times$ )	Stable (2–6 t/s)
Token generation	$7.8\to 12.3$ t/s ( $1.6\times$ )	Stable (2–6 t/s)
Stability	Continuous	160+ min, zero crashes
Without Moēsis	Functional but slow	Immediate OOM

Table 1: Moēsis performance on consumer hardware. The 6 GB configuration represents a laboratory stress test demonstrating system stability under extreme resource constraints, not a recommended operational configuration.

Figure 5: Mesh cross-KB routing: queries are routed to relevant knowledge bases via hierarchical fingerprint matching, then runtime structural discovery identifies emergent connections across separate graphs.

Capability	Existing Approaches	Noēsis
Cross-chunk context during extraction	Rare or limited	✓
Adaptive ingestion parallelism	Not observed	✓
Independently scalable inference backend	Configurable backends, not decoupled services	✓
MoE-aware selective quantization	Architecture-based or generic calibration; static	✓
Cross-KB routing ( $<2$ ms)	Not addressed	✓
Cross-KB structural discovery (runtime)	Not addressed	✓

Table 2: Capabilities of Noēsis versus common patterns in existing Graph-RAG approaches † .

Figure 6: Two-dimensional visualization of a knowledge graph generated by Noēsis (1,381 nodes, 3,740 edges, 82 communities). Node colors indicate Leiden communities; edge thickness reflects relationship strength.

System	EM	F1
GraphRAG † [ 1 ]	31.70	42.74
BGE dense + GPT-4o	47.60	60.36
Noēsis + GPT-4o	59.50	74.74
HopRAG † [ 20 ]	62.00	76.06
StepChain † [ 19 ]	66.70	79.50

Table 3: Multi-hop QA performance on HotpotQA (distractor, 1,000 questions). Baselines from [ 19 ] .

\dagger

denotes systems using GPT-4o for both graph construction and answer generation.

Configuration	EM	F1
BGE dense + GPT-4o	47.60	60.36
Noēsis + Gemma 4 E2B (2.3B)	47.20	60.30
Noēsis + GPT-4o	59.50	74.74

Table 4: Ablation: effect of answer-generation model on HotpotQA. The Noēsis graph (built by Qwen3.6-35B-A3B) is identical across both configurations.

Explore similar work

RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM
Jul 13, 2026Mikhail Komarov, Ivan Bondarenko, Stanislav Shtuka +5KG Construction LLM Reasoning with Graphs
1ITMO University · 2Novosibirsk State University · 3Far Eastern Federal University +1
ROZA Graphs: Self-Improving Near-Deterministic RAG through Evidence-Centric Feedback
Apr 8, 2026Matthew PenarozaRetrieval-Augmented Generation Graph Retrieval-Augmented Generation
GLM-RAG: Graph Language Models for Graph-Based Retrieval-Augmented Generation
Jul 30, 2026Maya Arseven, Anette Frank, Beni Egressy +2LLM Reasoning with Graphs Graph Retrieval-Augmented Generation
Institute of Computational Linguistics, Heidelberg University · Aleph Alpha Research

Jul 13, 2026J/K move · Enter open · S save

RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM

Mikhail Komarov, Ivan Bondarenko, Stanislav Shtuka, Oleg Sedukhin+4

1ITMO University · 2Novosibirsk State University · 3Far Eastern Federal University+1

KG Construction LLM Reasoning with Graphs Graph Retrieval-Augmented Generation Small Language Models Knowledge Representation LLM Information Extraction+1

Open paper AskPDF arXiv

Graph retrieval-augmented generation (GraphRAG) enhances large language models with structured knowledge, yet existing systems construct knowledge graphs in a single extraction pass, producing noisy entities and brittle retrieval. RAGU, an open-source modular GraphRAG engine, addresses this by separating extraction from consolidation: entities and relations pass through two-stage typed extraction, DBSCAN-backed deduplication, LLM summarization, and Leiden community detection. A key insight motivates a compact extractor: the skills an in-pipeline LLM needs - comprehension, extraction, reasoning over context - are language skills that grow only weakly with model size, unlike factual world knowledge. Accordingly, we train Meno-Lite-0.1, a 7B model optimized for language skills, which outperforms Qwen2.5-32B on knowledge-graph construction (+12.5% relative harmonic mean) and matches it on English GraphRAG tasks. On GraphRAG-Bench (Medical), RAGU retrieves the most complete context at every factoid level (evidence recall up to 0.84 vs.

\leq

0.76) and overtakes HippoRAG2 on synthesis tasks; on multi-hop factoid QA, the apparent HippoRAG2 advantage is shown to be largely an answer-format artifact. RAGU is installable via

\texttt{pip install graph_ragu}

, runs on a single GPU, and is released under MIT. The source code is publicly available at https://github.com/RaguTeam/RAGU, and the Meno-Lite-0.1 model can be obtained from https://huggingface.co/bond005/meno-lite-0.1.

Abstract

Explore similar work

RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM

ROZA Graphs: Self-Improving Near-Deterministic RAG through Evidence-Centric Feedback

GLM-RAG: Graph Language Models for Graph-Based Retrieval-Augmented Generation